27 commits

Author	SHA1	Message	Date
Danny Avila	6f45a9e32e	🔗 fix: Normalize MCP Tool Keys at Every Producer, Resolve Raw Names via Aliases (#14553) ... * 🔗 fix: Normalize MCP Tool Keys at Every Producer, Resolve Raw Names via Aliases Tool keys had two spellings that could diverge for any server whose name contains characters outside [a-zA-Z0-9_.-]: the tool cache (and registry inspector) built keys with the RAW server name, while runtime instances are named with normalizeServerName(serverName). Three code comments already asserted "tool keys embed the normalized server name" - no producer honored it. For a special-character server that meant: - definitions-only mode shipped raw def names the model echoed back, but the executor's tool map held the normalized instance name, so every call failed with "Tool not found"; - per-tool tool_options (defer_loading / allowed_callers / run_in_background / describe_intent) were persisted under raw keys that never matched the definition names the option passes resolve against, so builder settings were silently inert; - tool-key parsing against normalized candidate lists silently fell back to last-delimiter splitting, which mis-parses delimiter-bearing tool names. The reconciliation is one contract enforced in three moves: 1. PRODUCERS NORMALIZE. The tool cache (packages/api/src/mcp/tools.ts) and the registry inspector build keys with the normalized server name, matching the instance names MCP.js has always assigned. The builder's tool ids, agent.tools entries, tool_options keys, and definition names all flow from these keys, so every model-facing name now agrees. The cache STORE stays keyed by the raw config name. 2. CONFIG LOOKUPS RESOLVE ALIASES. New shared helpers in data-provider (buildServerNameAliases, normalizeMCPToolKey) map a parsed normalized name back to the raw config name that the registry, config maps, tool cache, and plugin-auth rows are keyed by. Applied in the definitions loader closure, handleTools grouping, createMCPTool's parsing fallback, getUserMCPAuthMap, and the MCP tools endpoint - matching both spellings so legacy raw keys keep resolving. 3. LEGACY DATA HEALS AT ONE BOUNDARY. initializeAgent rewrites raw-keyed agent.tools entries and tool_options keys to the normalized form (normalizeAgentToolKeys) before anything consumes them, so agents persisted under the old convention load their tools AND have all four per-tool options honored. Placeholder and server-pin tokens stay raw - they are config-identity references, not model-facing names. Servers whose names are already in the safe character set (the common case) produce byte-identical keys before and after; the fast path allocates nothing. Stale Redis-cached raw keys self-heal via the existing reconnect-on-missing path within one cache cycle. * 🧯 fix: Deterministic Alias Collisions + Raw Names in Definition Metadata Two review findings on the normalization contract: - Two configured server names that normalize to the same segment (e.g. 'Sales Force' and 'Sales:Force' -> 'Sales_Force') produce inherently ambiguous tool keys; the alias map silently resolved last-wins, so a tool selected from one server could execute against the other's config. buildServerNameAliases now resolves collisions to the FIRST configured name deterministically, and resolveMCPServerContext warns once per colliding pair per process so the operator can rename one server. A collision-resistant identifier would change every existing tool key, so detection + stable routing is the right treatment here; startup-time config validation can follow separately. - The definitions loader resolved parsed (normalized) server names to raw only inside the ToolService closure, while the definition metadata (serverName -> mcpRawServerName) kept the normalized value. Server instructions are keyed by raw config names, so a special-character server's instructions were silently omitted in definitions-only mode. loadToolDefinitions now takes rawServerNames, resolves the boundary against both spellings, and stores the RAW name in definition metadata - consistent with the instance path. * 🧯 fix: Heal Stale Caches, Skill Allowed-Tools, and Builder Selectors Three review findings on the normalization rollout, all in the transition class: - Stale cache entries (P1): the definitions-only loader treats the per-server tool map as authoritative and never reconnects on a per-key miss, so a pre-change raw-keyed Redis entry would make a special-character server's tools vanish for up to the cache TTL. getMCPServerTools now heals legacy raw-keyed entries to the normalized format at read time (keys and function names), covering every consumer with no coordinated invalidation; safe names return the map untouched. - Skill allowed-tools: a skill declaring a raw MCP key in allowed-tools bypassed the initialize-boundary heal (the union runs after it) and would neither dedupe against healed agent tools nor match the normalized tool map. The primes' allowedTools now pass through the same normalizeAgentToolKeys heal before unioning. - Builder selectors: matchesMcpServer and useVisibleTools parsed tool ids against raw server names only, so an attached special-character server rendered as an unselected orphan card. Both now accept the normalized spelling and resolve it back to the raw map key, keeping legacy raw ids working. * 🧯 fix: Fail Closed on Normalized Server-Name Collisions Escalation of the collision finding: a deterministic first-wins alias plus a warning still let the tools listing publish BOTH colliding servers, so a tool selected under the shadowed second server would silently execute against the first server's configuration (their model-facing keys are identical, so routing cannot ever distinguish them). - findShadowedServerNames identifies later-configured names whose normalized form an earlier different name claimed. - getMCPTools excludes shadowed servers from the published listing entirely (with a warn naming the collision), so their tools are never selectable - nothing ambiguous can be picked. - Server creation reserves both spellings: a generated slug may not collide with a raw config name OR the normalized form its tool keys would carry. Collision-resistant model-facing IDs remain out of scope: changing normalizeServerName's output would rewrite every existing tool key (agent documents, caches, instance names) for ALL servers to handle a misconfiguration that is now blocked from exposure instead. * ✅ fix: Dedupe Reserved Server-Name Spellings at Creation The reservation list appended normalized forms unconditionally, which duplicated every safe name (raw === normalized) and broke the route-level contract test pinning the exact list. Dedupe via a Set so safe names contribute one entry, while special-character names still reserve both spellings; adds the special-character reservation case. * 🧯 fix: Never Heal a Shadowed Server's Keys; Align Authorization Tie-Break Persisted references were the remaining collision vector: an agent or skill saved with the shadowed later server's raw key was HEALED into the shared normalized key, authorized through a last-wins map, and routed first-wins - authorized as one server, executed as another. - normalizeAgentToolKeys now refuses to rewrite keys of shadowed servers (findShadowedServerNames): rewriting would produce exactly the first server's key. Left raw, the key cannot match the normalized-keyed tool map and the tool fails visibly - broken beats misrouted. Covers agent.tools, tool_options, and skill allowed-tools through the shared heal. - filterAuthorizedTools (agents/v1.js) builds its normalized-to-raw map via the shared buildServerNameAliases instead of a last-wins Map constructor, so authorization resolves a colliding key to the SAME first server execution routes to. * 🧯 fix: Direct Identity Wins Over Aliases; Heal Client Forms and Degraded Contexts Four review findings on the normalization edges: - Alias hijack (P1): a user-DB server named exactly like an operator server's normalized form ('foo' vs YAML 'foo!') had its tools rerouted to the operator server by unconditional alias resolution. Resolution is now DIRECT-FIRST everywhere: the parsed name is tried as-is, and only when nothing resolves is it treated as a normalized spelling (definitions loader, handleTools grouping, createMCPTool fallback). buildServerNameAliases seats identity entries before derived ones so a literal name owns its slot regardless of config order, findShadowedServerNames and the collision warning derive from the same construction, and getUserMCPAuthMap fetches auth under both spellings so either owner finds its rows. - Builder double-match: a normalized name containing the delimiter ('foo mcp bar' -> 'foo_mcp_bar') also suffix-matched a server named 'bar', selecting both cards and making removal strip the wrong tool. matchesMcpServer now resolves the token ONCE against the full configured list (longest boundary, both spellings) when the caller supplies it; selection and removal share the resolution. - Builder legacy ids: an agent saved with raw-keyed ids showed its tools unchecked while the runtime heal kept them active, and selection updates never replaced the legacy entries. McpSection maps legacy raw ids to their current normalized ids when deriving and rewriting this server's selection. - Degraded context: a transient ensureConfigServers failure returned an entirely empty context, leaving normalized keys unresolvable for the request. resolveMCPServerContext now keeps the name lists (they derive from the config snapshot alone) and degrades only the lazy-init configs. * 🧯 fix: Collision Detection Sees Accessible Servers; Shadowed Refs Fail Closed End to End Round follow-ups on the collision design, all in the DB-server-visibility class: - The legacy-key heal detected collisions against operator-config names only, so healing could still produce a key that direct-first resolution routes to an invisible user-DB server. initializeAgent gains an optional getAccessibleMcpServerNames dep (wired through ToolService for controllers that mock it, directly elsewhere), consulted ONLY when a configured name needs normalization - zero cost for safe-name deployments. The heal then sees the full accessible set and skips shadowed servers' keys. - Wildcard and legacy raw tokens bypassed catalog filtering, letting a shadowed server's instances join a run under the same normalized names as the winner's. filterAuthorizedTools rejects tools of shadowed servers at authorization (its merged map sees DB + config), and handleTools skips them at execution. - The builder migrated only tool selection, not tool_options: legacy raw option keys showed disabled while the runtime honored them, and toggles could not clear them. McpSection now migrates option keys to the current normalized ids (existing normalized entries win). - A transient ensureConfigServers failure degraded to an EMPTY server context, leaving normalized keys unresolvable for the request. resolveMCPServerContext keeps the name lists (derived from the config snapshot alone) and degrades only the lazy-init configs. * 🧯 fix: Complete the Collision Audit at Every Gate; Safer Heal Semantics Round follow-ups hardening the collision audit: - Execution guards now consult the FULL accessible set: the caller's heal threads its already-fetched names through loadTools, and handleTools fetches them itself when a configured name needs normalization (never for safe-name deployments) - so a cross-tier collision (user-DB 'foo' vs operator 'foo!') fails closed at eager execution instead of joining the run under one normalized name. - Healing is SKIPPED when the collision audit cannot complete (transient lookup failure, or no dep): un-healed raw keys still resolve through the direct-first candidates, so skipping is safe while rewriting against an incomplete audit is not. - The audit lookup is gated on the agent actually carrying delimiter-bearing keys (tools, tool_options, or skill allowed-tools), so non-MCP agents never pay a registry round-trip even on specially named deployments. - normalizeAgentToolKeys gives the CURRENT (normalized) entry precedence when both spellings carry options, matching the builder's migration semantics instead of letting insertion order decide. - The builder's toCurrentToolId resolves entries boundary-exactly against every configured server (longest match, both spellings), so a raw suffix shared with a LONGER server name can no longer reassign that server's selection or options while another dialog is open. * 🧯 fix: Shared Collision Audit for Definitions Loading; Fail Closed on Audit Failure Round follow-ups closing the remaining audit gaps: - The definitions-only loader now consumes the same collision audit as eager loading: shadowed servers' entries (wildcards included) are dropped before definitions are emitted, so the default execution path can never resolve a shadowed server's normalized function name to another server. The audit names thread from initializeAgent's heal; the loader self-fetches only when a configured name needs normalization. - resolveCollisionAuditNames centralizes the audit-resolution policy (threaded set > self-fetch when needed > incomplete on failure), and BOTH loaders now fail closed under an incomplete audit: any normalization-sensitive reference (its own name needs normalizing, or it equals the normalized form of a configured special-character name) is skipped with a warning instead of being audited against operator names alone. isNormalizationSensitiveName lives in packages/api as a pure helper so test mocks use the real predicate. - normalizeAgentToolKeys collapses duplicate ids after healing (order-preserving): a document carrying both spellings converges on one key, never two instances with the same function name. * 🧯 fix: Thread the Audit Everywhere; Identity-Aware Alias Fallback Round follow-ups on audit plumbing: - The OpenAI-compatible and Responses tool loaders now forward the already-resolved accessibleMcpServerNames instead of discarding it, so the definitions loader neither repeats the registry lookup nor fails closed on a transient second lookup after the first succeeded. - The skill-only path threads its audit: when the baseline agent has no MCP keys but a primed skill's allowed-tools fetched the complete set, that set (not the operator-only list) reaches the loader, so the collision remains visible and the shadowed reference stays rejected end to end. - OAuth discovery iterates the collision-FILTERED tool list, so a request can no longer emit an OAuth prompt, wait out the connection timeout, and reconnect a server whose definitions were deliberately rejected. - The definitions loader's alias fallback is identity-aware: when the parsed name IS a known accessible server, a null tool fetch means temporarily unavailable (OAuth pending, missing user variables, disconnected) and no longer reroutes to the raw alias - previously the aliased operator server's definitions could be emitted under the unavailable DB server's names. * 🧯 fix: Legacy-Key Definition Lookup; Retain Audit for Deferred Execution - createMCPTool resolves tool definitions by BOTH spellings: the key as persisted plus the canonical normalized key built from the resolved server name. Assistants and direct tool calls persisted before the rollout bypass the agent-boundary heal and arrive with raw keys, while availableTools is now indexed canonically - previously every such call missed the index, burned a reconnect, and returned the unavailable stub permanently via the negative cache. - The initialized agent retains accessibleMcpServerNames (the COMPLETE collision audit this initialization resolved), buildAgentToolContext copies it into every per-agent tool context, and loadToolsForExecution threads it into the eager loader as bare options. Deferred/event-driven execution therefore reuses the snapshot instead of repeating the merged registry read - a transient failure there could fail-closed a tool the same turn already advertised from the successful first audit. - MCP.spec.js keeps @librechat/api pure helpers REAL (requireActual spread) so normalization paths are exercised rather than mirrored. * 🧯 fix: Parse Legacy Keys Against Both Server-Name Spellings createMCPTool's boundary candidates were normalized-only, so a legacy raw key whose server name contains the delimiter (foo_mcp_bar!) missed the suffix match and fell to the generic last-delimiter split - the canonical rebuild then produced a key that could never hit the index and the persisted call stubbed out. The candidate list now carries the RAW resolved name (and raw config names on the parse-only path) next to the normalized spellings. * 🧯 fix: Honest Audit Completeness; Shadowed-Server Form-Key Guard - resolveAllMcpConfigs tolerates ensureConfigServers failures, so the merged registry read can silently omit config-only servers while the audit still reported complete: true - a foo/foo! collision would go unseen and a persisted key could route to the wrong server. Both audit consumers now union the snapshot-derived raw config names back in (resolveCollisionAuditNames unions the caller's rawServerNames; the initializeAgent heal unions configRawServerNames), keeping the completeness label honest without an extra read: operator names come from the registry-independent config snapshot, user-DB names from the merged read that fails loudly into the existing incomplete path. - The client tool_options migration now mirrors the runtime heal's fail-closed rule for SHADOWED servers: when the dialog's server has lost its normalized slot to another catalog name, legacy raw keys stay raw instead of being rewritten onto the winning server's key, where a later save would apply the wrong server's per-tool settings. The dialog's own server joins the alias construction so a stale catalog map can't misread as a collision. * 🧯 fix: Heal Legacy Assistant MCP Tool Names on Save The assistants create/update controllers look tools up in the cached definitions by exact key, and the cache is now normalized-keyed - an assistant saved before the convention resubmits its raw-suffixed MCP name on every edit, so any save silently removed the tool. healMcpToolNames pre-heals the payload's tool list: a delimiter-bearing string that misses the cache resolves through the configured raw names (longest-suffix, boundary-exact) and rewrites to the normalized key only when that key actually exists in the cache. SHADOWED raw names stay raw and fail closed, mirroring the runtime heal; the config read happens only when a delimiter-bearing name actually misses, and read failures propagate (write path) rather than silently dropping tools. v2's update loop also stops re-reading the tool cache per iteration. * 🧯 fix: Full-Audit Shadow Set + Dedupe in the Assistant Key Heal - The assistant-save heal built its shadow set from operator config names alone, so a cross-tier collision (user-DB `foo` owning the normalized slot of operator `foo!`) looked unshadowed and the legacy key healed into the shared normalized name - which direct-first execution then binds to the DB server. The shadow set now comes from resolveCollisionAuditNames' full accessible audit, and an incomplete audit skips healing outright (every rewrite candidate is normalization-sensitive by construction, so raw-and-fail-closed is the only safe answer). - Healed string entries dedupe order-preserving: a payload carrying both spellings of the same tool collapses to one entry instead of expanding into duplicate function definitions the provider rejects.	2026-08-01 07:39:24 -04:00
Danny Avila	250aca375a	🔗 fix: Resolve MCP Tool-Key Boundary Against Configured Server Names (#14448) ... * fix: resolve MCP tool-name delimiter collision at invocation time MCP tool keys are identified internally as `${rawToolName}${mcp_delimiter}${serverName}` (delimiter `_mcp_`). Several call sites parsed this back apart with a naive `toolKey.split(Constants.mcp_delimiter)`, assuming the delimiter occurs exactly once. When the raw upstream tool name itself contains the delimiter substring - which happens whenever it's exposed through a gateway that prefixes aggregated tool names by server (e.g. a gateway's own "gitlab-get_mcp_server_version" for GitLab's "get_mcp_server_version" tool) - the combined key has the delimiter more than once. `.split()` then produces more than two segments, and destructuring `[toolName, serverName]` silently keeps only the first two, yielding a bogus server name that matches no configured server. Tool listing still worked (a different code path builds keys directly without re-splitting), but invocation failed with `Tool {name} not found`, and `filterAuthorizedTools` rejected such keys outright as malformed. Add `splitMCPToolKey`, which splits on the *last* occurrence of the delimiter instead: the server-name half is always LibreChat's own normalized suffix (guaranteed not to contain the delimiter), while the raw tool-name half is untrusted and may legitimately contain it. This matches `.split()`'s result whenever the delimiter occurs once, and correctly resolves the collision case. Update the four call sites that parsed this manually (`handleTools.js`, `MCP.js`, `mcp.js` controller, `filterAuthorizedTools` in `v1.js`) plus one in the client (`useVisibleTools.ts`) to use it. Fixes #14440 * fix: resolve MCP tool-key boundary against configured server names splitMCPToolKey moves to librechat-data-provider so the client and backend share one parser, and takes the configured server names when the caller has them: the longest name the key actually ends with wins, which is exact. Position alone cannot identify the boundary because both halves may contain the delimiter. lastIndexOf alone fixes gateway-prefixed tool names but regresses servers whose own name contains it, which ToolService.spec.js already covered; the last-delimiter path now only serves as the fallback for callers with no configured set. Also converts the remaining first-occurrence parsers that the delimiter fix missed - mcp/auth.ts (custom user vars silently unresolved), mcp/oauth/events.ts, agents/initialize.ts, and the three client parsers that labelled tool calls with the wrong server. * fix: keep client tool-call labels on first-delimiter parsing The three client parsers had deliberate, tested first-delimiter semantics (ToolCall.test.tsx asserts the full server name for 'foo_mcp_bar' and the synthetic 'oauth_mcp_server' call), and the client has no configured server list in scope to resolve the boundary exactly, so they are left as they were. Threads the configured names into the event-driven definition loader so it resolves the same boundary as the authorization filter that admits the key, and documents the one case that stays undecidable without provenance. * fix: resolve tool-key boundary against all configured servers resolveConfigServers only returns lazily-initialized config overrides - ensureConfigServers skips unmodified YAML servers - so on a stock deployment the known-name list was empty and suffix resolution never engaged. Adds resolveMcpServerNames, which keeps every configured server in the normalized form tool keys carry, and uses it at the loading, auth-map and definition sites. Background-tool eligibility now resolves against all configured names before testing ephemeral membership, so a non-ephemeral server whose name ends in an ephemeral one is no longer misclassified, and useVisibleTools resolves against the server map it already receives. * fix: use resolved server provenance and one app-config read createMCPTool now uses the serverName loadTools already resolved for the key and only parses as a fallback, so an unmodified YAML server whose name contains the delimiter no longer resolves to the wrong server for auth, reconnection and callTool. resolveMcpServerContext derives config servers and all configured names from a single getAppConfigForRequest, replacing two independent lookups on the chat startup path, and degrades to empty like resolveConfigServers instead of aborting tool loading when the config lookup fails. * chore: drop unused resolveConfigServers import * fix: forward server provenance on the all-tools path and read config once createMCPTools builds each toolKey from the server name it already has but did not forward it, so the sys__all__sys path re-derived it by parsing and bound an unmodified YAML server whose name contains the delimiter to the wrong auth and invocation context. loadAgentTools now resolves the MCP server context once and threads it into loadTools, replacing the second app-config read it had introduced on the non-event-driven chat startup path. * fix: carry resolved MCP server name through tool classification definitions.ts resolves the server for each key and then dropped it when building loadedTools, so buildToolClassification re-derived it with a last-segment split and recorded 'Workspace' for a server configured as 'Google_mcp_Workspace'. The resolved name now rides along on the tool instance and classification prefers it over re-parsing. * fix: consume carried server name when extracting MCP servers extractMCPServers re-derived the name with a last-segment split, so a server configured as Google_mcp_Workspace resolved to Workspace and its instructions were silently omitted. Prefers the name carried on the tool definition instance, falling back to the split. * fix: fail closed on ambiguous MCP keys when persisting server names Persisted mcpServerNames grant agent-scoped access to a DB server by name (ServerConfigsDB.getAccessibleServers), so a wrong guess exposes an unrelated server to everyone who can view the agent. The last-segment split turned search_mcp_Google_mcp_workspace into 'workspace'; such keys were previously rejected outright at agent save, so admitting them opened this path. Derives a name only from unambiguous single-delimiter keys. This is #12250's guard moved to the boundary it was actually protecting, instead of blocking tool admission. * fix: keep DB server access for multi-delimiter tool keys The fail-closed guard was wrong for the case this PR exists to fix. This index only grants DB-backed servers, and DB names are slugs that cannot contain the delimiter (generateServerNameFromTitle strips underscores), so the trailing segment is always the real server for them - dropping it cost every consumer of a gateway-prefixed tool their shared-agent access. Also gates the MCP server-context lookup on the filtered MCP set, so an agent with no MCP tools no longer pays an app-config read on startup. * fix: resolve tool-call display names without breaking OAuth calls The display parsers could not use the shared boundary parser because their tested behavior depends on first-delimiter semantics. That constraint only applies to synthetic MCP OAuth calls, whose tool half is always exactly 'oauth', so everything after the first delimiter is the server even when the server name carries one. splitToolCallName special-cases that form and defers to splitMCPToolKey for real tool keys, so a gateway-prefixed tool now renders its own name and server while oauth_mcp_foo_mcp_bar still resolves to foo_mcp_bar. * fix: persist resolved MCP server provenance on agents Deriving mcpServerNames from the tool key cannot tell a config server's trailing segment from a real DB server name, so a config server named a_mcp_b indexed an unrelated DB server b and shared the agent's viewers into it. Neither string rule works: the suffix guess exposes, and failing closed drops legitimate DB access for gateway-prefixed tools. filterAuthorizedTools already resolves each tool's server against the merged registry config, so it now collects those names and create, update and duplicate persist them. No extra registry queries: the update path unions the newly resolved names with what the agent already had, and duplicate replaces the copied list rather than inheriting the source's servers. Display parsing also takes the configured names, so a real tool call on a delimiter-bearing server renders the right server and icon. * test: teach MCP hook mocks about useMCPServerNames Three specs mock ~/hooks/MCP with a hand-listed factory, so adding the hook to ToolCall made useMCPServerNames undefined under test and every render threw. Returns a stable array so the mock cannot perturb render counts. * fix: rebuild agent MCP server index from surviving tools Unioning the prior names kept a server indexed after its last tool was detached, so viewers of a shared agent retained agent-scoped access to it. The index is now rebuilt from the tools that survive the edit: a prior name carries forward only while some retained tool still resolves to it, using the agent's own persisted names as the candidate set, and the rebuild runs on any tool change rather than only when a new MCP tool is added. * fix: keep duplicate indexes on registry fallback and harden the oauth split Duplication blanked mcpServerNames when the registry was unavailable, because filterAuthorizedTools grandfathers the source's tools without resolving them - the copy kept tools it could no longer resolve. Source names now carry forward for the tools that still point at them. splitToolCallName also treated any oauth_mcp_ prefix as a synthetic OAuth call, so a genuine upstream tool by that name resolved to the wrong server. A configured server name now decides when one matches, since a real key always ends in its server, and the prefix only breaks ties for unconfigured servers. * fix: thread configured server names through display parsing parseToolName and getMCPServerName resolved context-free, so a configured server whose name contains the delimiter showed the wrong server in grouped tool summaries and subagent tool labels, and stacked icons missed its entry in the icon map. Both take the configured names now, supplied by the components that render them. Adds the hook to SubagentCall's mock factory: the spec renders the real component, so an unmocked useMCPServerNames would reach the query with no provider. * test: cover the auth-map boundary, server provenance and context fallback Adds regression coverage for three behaviors this PR changed that no test exercised: customUserVars resolving under the right plugin key for a gateway-prefixed tool name (the failure that made these tools loadable but unusable), the resolved server name reaching createMCPTool instead of being re-parsed, and resolveMcpServerContext degrading to empty rather than aborting tool loading when the config lookup fails. Each was checked against a mutated source to confirm it fails when the behavior is broken. * fix: normalize server-name candidates and cover the boundary guard Tool keys embed normalizeServerName's output while the config is keyed by the raw name, so callers passing raw keys never matched a server whose name needs normalizing and silently fell back to the last delimiter. filterAuthorizedTools now maps normalized names back to their config key, and createMCPTool normalizes its candidates. Adds the cases an audit found surviving mutation: a configured name that is a bare but not delimiter-aligned suffix must not match, an empty candidate list behaves as no list, and splitToolCallName still falls back to the oauth prefix when a list is supplied but nothing in it matches. * fix: keep resolved server names when a non-owner retains MCP tools The shared-agent path keeps an agent's existing MCP tools verbatim but supplied no mcpServerNames, so persistence re-derived them and reduced a configured server like Google_mcp_Workspace to Workspace - which ServerConfigsDB then treats as a DB server, granting the agent's viewers access to an unrelated one. Carries the existing resolved names across instead, and clears the index on the owner path where every MCP tool is removed. * fix: preserve resolved MCP names for every tools update extractMCPServerNames was reachable from any caller that writes tools without mcpServerNames - the Action edit path does exactly that - so a configured Google_mcp_Workspace was reindexed as Workspace and ServerConfigsDB granted shared-agent viewers an unrelated DB server by that name. updateAgent now rebuilds the index from the agent's own resolved names: one carries forward while a retained tool still resolves to it, and only keys matching none of them fall back to derivation. Callers are safe by default rather than by remembering to pass the set. normalizeServerName moves to librechat-data-provider so the client can match its candidates against tool keys, which embed the normalized form; the icon map is keyed the same way since it is looked up with a parsed server name. * refactor: move MCP context resolution into packages/api New backend logic belongs in the TypeScript workspace per CLAUDE.md, with /api kept to a thin wrapper. resolveMCPServerContext now lives in packages/api/src/mcp/context.ts and takes ensureConfigServers by injection, since the registry accessor is still legacy-only; the /api function is reduced to loading the request app config and translating failures into the empty degrade it already promised. * test: teach the MCP service mock about resolveMCPServerContext The spec mocks @librechat/api with a hand-listed factory, so moving the resolver into that package left it undefined and the wrapper degraded into its own catch, returning empty config servers. The stub mirrors the real resolver so these tests still cover what the wrapper owns - loading the request config and degrading on failure - while the resolution logic is unit-tested in packages/api. * fix: only persist an authoritative MCP server index on update Assigning the resolved set unconditionally pinned the index to [] whenever nothing authoritative was available - a legacy agent holding MCP tools with no stored mcpServerNames - which suppressed updateAgent's derivation and stripped agent-scoped access to its DB-backed server. The field is now supplied only when the result is authoritative: names were resolved, or no MCP tool survives so the index genuinely is empty. The retained-tools branch likewise leaves it unset when the agent has none stored. --------- Co-authored-by: Jens Schumann <schumajs@gmail.com>	2026-07-27 14:45:38 -04:00
Danny Avila	d8427ffc5e	🛂 test: Cover Tool Approval Workflows End to End (#14427) ... * test: cover tool approval workflows end to end * fix: preserve tool approval state across resume * fix: preserve agent context in mock stream responses * fix: preserve nested approvals in collapsed groups	2026-07-26 21:58:25 -04:00
Danny Avila	988a14a405	🙋 feat: ask_user_question - agent-initiated questions with durable pause/resume (#14139) ... * feat: ask_user_question tool — agent-initiated questions with durable pause/resume The HITL runtime merged in #13942/#14024/#14025/#14123 already ships the full ask_user_question lifecycle (payload-agnostic handleRunInterrupt, resume validation via mapAskUserAnswer, reconnect rehydration, and the client question card) — but nothing ever raised the interrupt. This adds the producer: - packages/api/agents/hitl/askUserQuestionTool.ts: LLM-callable tool whose func calls the SDK askUserQuestion() helper (LangGraph interrupt() from the tool body); zod schema with length caps mirroring AskUserQuestionRequest, plus a JSON-schema twin for the schema-only registry - Registration: agentToolDefinitions, manifest.json (Tools dialog, admin filteredTools/includedTools kill switch), basicToolInstances, handleTools constructor branch - run.ts gating: checkpointer now attaches for hitlCapable runs whose agents carry the ask tool even with the tool-approval policy disabled (the interrupt needs only durability, not humanInTheLoop/hooks); the tool is stripped fail-closed from non-HITL callers (OpenAI-compat/Responses) and subagent child configs; excluded from eager event execution (interrupts must be raised inside the Pregel task frame) - resume.js: 16k length cap on the answer wire field - e2e (real Run + FakeChatModel + LazyMongoSaver + supertest resume): tool-body interrupt pauses durably with NO approval policy, answer round-trips as the ToolMessage content, tool body re-runs once on resume, sequential questions re-pause * fix: adversarial-review findings — in-graph execution, orphan prunes, endpoint scoping, real kill switch Pre-PR multi-agent review confirmed 5 defects in the initial commit; all fixed: 1. CRITICAL — the tool never paused on the real agents endpoint: production loads tools definitions-only, flipping the SDK ToolNode to event-driven dispatch, and the host ON_TOOL_EXECUTE handler runs outside the Pregel task frame (under runOutsideTracing), where interrupt() throws and becomes an error ToolMessage. Reworked: the ask tool never rides toolDefinitions/ toolRegistry — on HITL-capable top-level agents a real instance is supplied via AgentInputs.graphTools (agents#289, requires @librechat/agents > 3.2.57), the SDK's in-graph direct-tool seam; new production-shape e2e pins the event-driven mode end to end. 2. CRITICAL — ask-only runs left orphaned interrupted checkpoints (silent context duplication on every later turn): both orphan prunes were gated on toolApproval.enabled. The pre-turn prune now also fires for ask-capable agents (exported agentRequestsAskUserQuestion), and the abort-route prune fires when the aborted job carries a pendingAction. 3. MAJOR — self-spawned subagents bypassed the strip (self config resolves from the parent's _sourceInputs): fixed SDK-side (buildChildInputs clears graphTools) and the tool is now never present on child surfaces host-side. 4. MINOR — the manifest entry leaked into the Assistants tools dialog and the legacy plugins endpoint, where tools execute with no run to pause: new agentsOnly manifest flag, scoped out of both listings. 5. MINOR — filteredTools/includedTools only hid the tool from the dialog: now enforced at run build (strip + no checkpointer), making the admin filter a real kill switch for already-saved agents. * chore: update @librechat/agents dependency to version 3.2.58 in package-lock.json and package.json files * fix: reject agents-only tools at assistant create/update (Codex round 1) The tools-dialog scoping keeps ask_user_question out of the assistants LISTING, but the v1/v2 create/update handlers resolve arbitrary posted tool strings from the shared getCachedTools map — a REST client or stale saved payload could still attach it, and the assistants runtime executes tools with no run to pause, so every call would error. New isAgentsOnlyTool(tool) (manifest-driven, handles string and function-object shapes) drops such tools with a warn at all four resolution sites (v1+v2, create+update). * fix: offset resumed-run content indices past the pre-pause seed A resumed run rebuilds the graph from the checkpoint, and the fresh graph numbers content indices from its own empty contentData — starting at 0. The resume path seeds the (also fresh) content aggregator with the pre-pause parts at exactly those indices, so the resumed model turn collided with the seed: type-matching parts silently MERGED (post-resume text appended into a pre-pause text block), and type-mismatching parts (a reasoning/think part at index 0 — any Anthropic reasoning agent) dropped EVERY delta with 'Content type mismatch', losing the entire post-resume output from the live stream and the saved message. Latent since #13942 — tool-approval resumes corrupt content the same way (probe-verified); it surfaced now because ask_user_question makes pausing a first-class flow and reasoning models make the loss total. - createContentIndexOffsetHandlers(handlers, offset): wraps ON_RUN_STEP (the single point where a content index enters the pipeline — deltas resolve through the aggregator's stepMap) and ON_AGENT_UPDATE's inline index; every other handler passes through by reference. Probe-validated: resumed output now lands as a new part after the paused tool call. - resumeCompletion wires it with offset = seedContent.length. - logToolError: a GraphInterrupt unwinding out of a tool body is the HITL pause working as designed — no longer logged as a Tool Error. * fix: unblock live streaming of the resumed segment after an answer With resume indices now ABSOLUTE (server continues after the pre-pause parts), the synthetic ask-user-question card was squatting on exactly the index the resumed segment streams into: applyAskUserQuestion appends the card at the end of the message content, so on the answering device every incoming part at that index was blocked and nothing rendered between the answer submission and the finalize replacing the message. removeAskUserQuestionPart(message, actionId) strips the pause-scoped card on successful answer submission (useResumeSubmit onSuccess) — the durable record of the Q&A is the ask_user_question tool call itself. Pure helper + specs; same-reference no-op when nothing matches. * fix: displace the synthetic question card in the streaming content writer The store-level strip on answer submit wasn't enough: the SSE step handler keeps its own in-flight copy of the streaming message, so on the answering device the synthetic ask-user-question card still occupied the ABSOLUTE index the resumed segment streams into — every delta warned 'Content type mismatch' (existing ask_user_question vs incoming text) and nothing rendered between the pending_action and finalize. Displace the card inside updateContent when any real part claims its slot — the same displacement pattern as the OAuth prompt part directly above it. Covers the streaming handler's own copy, reconnecting tabs, and other devices; once real content streams, the pause is over by definition. Spec drives a runStep + text delta into the card's index and pins: no mismatch warn, card gone, text rendered. * feat: dedicated UI + durable data for completed ask_user_question calls The completed ask call rendered as a generic tool card labeled 'Cancelled' with raw (and empty) JSON args. Two layers fixed: Data: the saved tool_call part had args:'' and no output — streamed arg chunks carry no tool name so the aggregator drops them (normal tools recover via the completion event, which never fires for a tool that interrupts mid-execution and resumes on a rebuilt run with no step id). The resume controller now stamps the paused ask part with the pendingAction's authoritative question as args and the user's answer as output (attachAskUserQuestionAnswer — pure, targets the newest unanswered ask part, so sequential questions each keep their own answer). UI: Part.tsx routes ask_user_question tool calls to AskUserQuestionCall — a compact Q&A record ('Asked a question' header, question, description, 'You answered: <label>' preferring the picked option's label, or 'No answer was given' for an abandoned pause) instead of the generic card. New i18n keys; parseAskUserQuestionArgs degrades to null on malformed model args. * fix: single question UI per pause + immediate answer display Two live-turn issues with the new durable Q&A card: 1. Duplicate question on ask: during a live pause the message carries BOTH the ask tool_call part (now rendered by AskUserQuestionCall, showing a misleading 'No answer was given' while paused) and the synthetic interactive card. The durable card now defers while the turn is live and unanswered (isSubmitting) — the interactive card owns the question UI until it's answered; an abandoned pause still shows its no-answer state once the turn settles. 2. 'No answer was given' after answering: the server stamps the answer onto the part at resume seed, but the client only received that at finalize. No stream emission needed — the client knows the answer it just submitted: resolveAskUserQuestionPart (replacing the plain strip on submit success) removes the synthetic card AND stamps output/progress onto the newest unanswered ask tool_call, seeding args from the synthetic part's question when the streamed args were lost — mirroring the server-side attachAskUserQuestionAnswer, so the Q&A record shows the answer the moment the user submits. * fix: keep the Q&A record visible while the resumed segment streams The optimistic output stamp lives in the message store, but the SSE step handler evolves its own cached copy of the streaming message (created at turn start) — the first resumed event overwrites the store with that copy, wiping the stamp, so the Q&A card blinked out during streaming and only returned at finalize. Render-layer fallback instead of fighting the handler's copy: submitted answers are recorded by ask tool_call id when resolveAskUserQuestionPart stamps the part, and AskUserQuestionCall reads the recorded answer whenever the part's own output is missing — the record survives any message-copy churn until finalize delivers the server-stamped part. * feat: present Ask User as a native builtin in the tools dialog It ships with the app and pauses the run like a first-class feature, so it belongs with the builtins (Run Code, Web Search, Memory, ...) rather than in the third-party plugin list — while its mechanics stay exactly a plugin's: - BuiltinId += 'ask_user_question' (documented exception: a native TOOL, not a capability; selection reads agent.tools, the toggle emits tool-add/remove patches instead of a capability field) - buildCatalog surfaces it as a builtin gated on the same signals as before (tools capability on + the server lists the plugin, i.e. not admin-filtered) and skips it in the plugin loop so it never double-lists - On-theme icon: lucide MessageCircleQuestion in a teal chip via the builtin icon map, matching the other native entries; the bespoke purple SVG and the manifest icon field are gone - i18n'd name/description keys like the other builtins * feat: composer popover for answering questions (mentions-style) Answering moves to the composer, matching the existing mentions/prompts popover pattern: while an ask_user_question pause is live, a popover anchors above the textarea with the question as its header, numbered option rows (hover/click, or ↑/↓ + Enter from the empty composer), and an × to dismiss. The main textarea doubles as the free-form answer — its placeholder flips to 'Something else...' and form submit routes the text to the paused run as the answer instead of starting a new turn. Dismissing (× or Escape) restores normal sends; the inline transcript surfaces stay as before (interactive card while paused, durable Q&A record after) so the question remains visible in history. - findLiveAskUserQuestion (pure, spec'd): newest unanswered synthetic part across the conversation IS the popover signal — applied on on_pending_action, stripped on answer submit, so visibility tracks the pause lifecycle with no extra state - useLiveAskUserQuestion hook shared by the popover and ChatForm; dismissals in a recoil atom so both react - popover only mounts on the primary composer (index 0), mirroring QuoteButton * feat: number-key selection + return glyph in the question popover Pressing 1-9 in the empty composer picks the matching option directly, mirroring the numbered row chips; the highlighted row shows a return-key glyph as the Enter affordance. Same empty-composer guard as the arrow keys — typing a free-form answer is never intercepted. * refactor: first-class composer answer mode (useAskAnswerMode) Replaces the bolted-on integration (inline onSubmit interception + raw capture-phase keydown listeners on the textarea ref) with a single hook that owns the whole answer mode: live-question derivation, dismissal + highlighted option (shared recoil state), option selection, free-form submit routing (submitText returns whether it consumed the submission), and keyboard handling (handleKeyDown returns whether it consumed the key, composed ahead of the textarea's normal handler — no more addEventListener). The popover is now pure rendering off the hook; ChatForm wires placeholder, onKeyDown, and onSubmit through the same instance. Deliberately scoped to the composer rather than useSubmitMessage: starters/prompt-commands keep new-turn semantics (and the existing job-replacement behavior while paused). * fix: Codex round 2 — inline answer input, approval exemption, pause-time args F1 (composer submit unreachable while paused — isSubmitting keeps Stop shown and useTextarea eats Enter): redesigned around it, borrowing Claude Code's AskUserQuestion semantics. The popover now owns free-form input via an inline 'Other' row (numbered last, 'Something else…'), with select-then-confirm rows (click/arrows/digits highlight; Submit ↵, Enter, or double-click fires; Skip dismisses). The composer returns to being a plain composer — no placeholder swap, no submit interception; Stop keeps meaning stop. F2: ask_user_question is exempt from the tool-approval prompt unless the admin explicitly lists it (allow/ask/deny all win) — approving the right to ask a question was a pure double pause; the tool is side-effect-free. F3: the question is stamped onto the paused ask tool_call's args at PAUSE time (attachAskUserQuestionArgs in handleRunInterrupt), so abandoned/expired/ stopped turns persist with the question intact and the record card can render it — previously only the answer-resume path stamped args. * fix: fold model-supplied 'Other' options into the inline free-form row The model can generate its own catch-all option ('Other (type your own)', value 'other'), duplicating the popover's built-in free-form row — two other-ish rows, one pickable as a literal answer. Two layers: - Tool description now tells the model NOT to include catch-all options (the answer UI always offers free-form input on its own) - splitOtherOption (pure, spec'd) folds a catch-all option that arrives anyway out of the choice rows and uses its label as the inline input's placeholder — conservative match (value 'other', or a label reading as a free-form invitation), no false positives on real choices * fix: single question surface + clean free-form-only popover Two live-pause confusions: (1) the inline transcript card and the composer popover both rendered — the card now defers while the popover is up for its action, returning as the fallback surface when the user dismisses the popover (and in contexts without a ChatContext, where the popover can't exist); (2) an options-less question showed a pointless numbered '1 Something else…' row — free-form-only questions now render the inline input alone, with the 'Type your answer…' placeholder (a folded model 'Other' label still wins). * feat: the composer is the free-form answer box (like the main chat input) While a question pause is live, the main chat textarea composes the free-form answer — placeholder swaps to 'Something else…' (or a folded model 'Other' label), Enter with text submits the answer through answer-mode key handling (composed BEFORE useTextarea's submitting-lock, so the lock can't swallow it), and the Stop button swaps to Send (enabled despite isSubmitting) per the select-then-confirm design. The popover slims to the question header, numbered option rows, and Skip/Submit — its inline input is gone since the composer owns free-form now. Dismissing the popover restores normal composer semantics (Stop button, normal sends). * fix: Codex round 3 + real Skip semantics - Skip now ANSWERS instead of hiding UI (danny): it resumes the run with a decline notice ('The user chose not to answer this question.') so the model moves on — a client-side dismiss left the run paused until expiry, a hung turn. × / Escape remain pure dismiss (switch to the inline card surface). - P1 (resumed approval tool indices): resumed tool_calls steps whose tool_call id matches a seeded UNRESOLVED part now rebind to that seeded slot instead of offsetting — the original part resolves in place (output attaches) and no duplicate appears; message steps keep the offset, so the text-loss fix stands. createContentIndexOffsetHandlers now takes the seed array; resolved seeded calls are not rebind targets. - P2 (stale selection across questions): selection state resets when the live actionId changes; the vestigial inline-Other state ('other' selection + text atom) is gone — the composer owns free-form. - P2 (Redis abort path loses the args stamp): the abort route re-stamps the question onto the ask tool_call in the reconstructed abort content, so a Stop-abandoned question persists with its question intact. - P2 (malformed args crash): parseAskUserQuestionArgs normalizes untrusted shapes (options: {} / non-string entries) instead of throwing in render. * feat: free-form hint in the question popover footer Left-aligned in the footer row (opposite Skip/Submit): 'Or type your answer below' — points open-ended answering at the composer, whose placeholder already reads 'Something else…'. * feat: preserve composer drafts across the answer-mode swap The answer phase gets its own draft key (ask-answer:<actionId>), passed as a draftId override into useAutoSave — the key change itself drives the existing save/restore machinery, so the conversation draft (or mid-run PENDING draft) is stashed when a question pause takes the composer and restored once the user answers, skips, or dismisses. Ask keys are exempt from the PENDING migration branch, which would otherwise move-and-delete the stashed draft. A half-typed answer survives reload/navigation while its question stays live. Answer submission (option pick, free-form, skip) resets the composer via a new non-throwing useOptionalChatFormContext, so the swap-back restores into an empty box even outside ChatView-less render contexts (Share/search). * fix: rebind resumed steps for ALL seeded tool call ids The resume controller pre-stamps the user's answer onto the seeded ask_user_question part, so the unresolved-only rebind predicate treated it as settled and shifted the tool's re-run step to a fresh offset slot, leaving a duplicate ask record in streamed/saved content. Tool call ids are provider-minted per call: a resumed step bearing a seeded id can only be the interrupted batch re-executing, so rebinding every seeded id is always correct. * feat: popover UX round 4 — clickable hint, collapse, click-submit, multiSelect - Footer hint is a button that focuses the composer; reads 'Type your answer below' (no 'Or') when the question has no options. - Collapse (chevron) hides the popover WITHOUT closing the pause: answer mode stays live (placeholder, Enter routing, draft key), the chat card renders the question with a ChevronUp affordance to re-expand. x remains dismiss. - Single-select options submit on a single click; the Submit button renders only for multi-select. - multiSelect end-to-end: tool zod schema + JSON definition twin, wire type, client parse, popover check-chips, card toggles, record-card label mapping; answer = option values joined ', '; composer Enter and the multi Submit button both fold free-form text in with the checked values. - Hardening from adversarial review: in-flight status guard on every submit path (no duplicate resumes on double-click), popover locks while submitting, collapsed mode disarms invisible digit/arrow steering, the card shares the hook's checked state while the pause is live, the card folds catch-all 'Other' options, record mapping is all-or-nothing to avoid phantom labels, composer resets only when its text was consumed or the draft machinery will restore the stash. * feat: ask_user_question in model specs and ephemeral agents A librechat.yaml modelSpec can now equip the tool the same way it equips webSearch/executeCode/fileSearch/memory: modelSpecs: list: - name: my-spec askUserQuestion: true loadEphemeralAgent pushes the tool name when the spec flag (or the ephemeralAgent request flag, wired for parity) is set; everything downstream is the existing persisted-agent machinery — createRun's hitlCapable gating, graphTools injection, checkpointer attach, subagent strip, and the admin filteredTools/includedTools kill switch all apply unchanged. * feat: tense-aware Q&A record label (Asking / Asked) Shorten the record card header per feedback: 'Asking' while the question is still unanswered (abandoned/awaiting), 'Asked' once answered — replacing the single 'Asked a question' label. * fix: Codex round 4 — added-agent ask parity + preserve answer on failed resume F1 (added.ts): mirror loadEphemeralAgent's ask_user_question branch in the added-agent loader so a model spec's askUserQuestion flag (or the ephemeral request flag) equips added top-level agents too, matching execute_code / web_search / memory. Two load.spec cases added. F3 (composer): submitAskAnswer now takes an onSuccess callback and useAskAnswerMode defers clearing the selection/composer until the resume is accepted. A failed resume (16k answer-cap 400, expired action, network error) leaves status re-answerable, so wiping the composer up front lost the user's only copy of a free-form answer; now it survives for trim/retry. (F2 — a claimed Tools-capability bypass — was verified NOT reproducible: agentRequestsAskUserQuestion matches only loaded instances/toolDefinitions/ toolRegistry, all capability-filtered; a raw tools string has no .name and never triggers the install. Replied on-thread with the probe evidence.) * fix: Codex round 5 — expired question exits answer mode so its message shows An expired question (e.g. resume returns the stale-action 409) previously left the popover open with locked controls and no explanation, because the chat card — which carries the only 'this action expired' message — was suppressed by the popover-open guard. Treat 'expired' as no longer active: the popover closes, the composer reverts to normal, and the card becomes the sole surface and renders the expired message. 'error' stays active (retryable). * feat: group ask_user_question calls as their own category A homogeneous group of ask_user_question tool calls now reads 'Asked N questions' (present tense 'Asking N questions' while the turn streams) with a question glyph and no raw-name suffix — mirroring the subagent 'Ran N agents' category treatment, instead of 'Used N tools — ask_user_question'. Mixed groups keep 'Used N tools' but humanize the suffix to 'Question' and show a question icon for the ask entries (TOOL_FRIENDLY_NAME_KEYS + ToolIcon map). A group only forms at count >= 2, so the plural is always grammatical. Three ToolCallGroup.test cases cover homogeneous label/icon/suffix, present tense while streaming, and the mixed-group fallback. * fix: Codex round 6 — composer submit lock + abort stamp before emit F7 (composer status lock): the ask submit status lived on ApprovalContext, a React context mounted only around message content (ContentParts). The PRIMARY answer surface — the composer in ChatForm — renders outside it, so useApprovalContext returned the inert FALLBACK: status was always 'idle', setStatus a no-op. The in-flight double-submit guard (round 4) and the expired-exits-answer-mode fix (round 5) therefore never engaged for the composer. Move ask submit status to a global Recoil atom (useAskSubmitStatus) read/written by the composer, the popover, and the card alike, so a fast double-click/Enter is actually blocked and expired/error surfaces on every surface. Tool-approval status stays on the context (unchanged). F5 (abort stamp before emit): the abort route re-stamped a paused ask_user_question's args AFTER GenerationJobManager.abortJob had already emitted the final SSE from the unstamped content, so a Redis/cross-replica Stop left the live client showing an empty question until reload. abortJob now takes an optional transformAbortContent applied to the persistable content BEFORE the final event is built (and returned), so the live client and the saved message agree. New abort.spec case + updated call assertions. * feat: gate ask_user_question behind its own agent capability Add a first-class AgentCapabilities.ask_user_question (in defaultAgentCapabilities, on by default) so admins can enable/disable questions independently via endpoints.agents.capabilities, exactly like execute_code / web_search — not lumped under the generic tools capability. - ToolService: both filteredTools predicates (definitions-only and instance loaders) gate ask_user_question on checkCapability(ask_user_question) before the generic tools fallthrough. When off, the tool is dropped from toolDefinitions/toolRegistry, so run.ts's agentRequestsAskUserQuestion (which keys on the loaded surface) declines to install it and attach a checkpointer — the capability is enforced end-to-end at the loader, no run.ts change needed. - Tools dialog catalog: surface the ask builtin under its own capability rather than the generic tools one, so the UI matches the backend gate. - Tests: ToolService capability on/off filtering + defaults membership; catalog builtin visibility keyed on the dedicated capability. * style: sort imports in ToolCallGroup.test (CI import-order gate) * fix: Codex round 7 — surface ask-answer errors in the open popover A failed answer submission (16k reject, network error) sets the ask status to 'error', which — unlike 'expired' — deliberately keeps the question active and retryable. But the chat card that renders the error message is suppressed while the popover is open, so a composer/popover answer failed silently. Expose an 'errored' flag from useAskAnswerMode and render a warning line (com_ui_ask_answer_error) in the popover, so the user gets feedback and retry guidance without having to collapse/dismiss. It clears automatically on retry (status flips to 'submitting'). * fix: Codex round 8 — respect IME composition before submitting answers handleComposerKeyDown runs before useTextarea's composition guard, so with a CJK/IME keyboard the Enter that commits an in-progress composition was being intercepted and submitting the partial answer (and the composition buffer can leave value empty mid-compose, mis-triggering digit/arrow steering too). Bail at the top when composing — nativeEvent.isComposing, or key==='Process' / keyCode===229 for Safari's inconsistent reporting — mirroring the existing composer guard so the character commits normally. * chore: update `@librechat/agents` to v3.2.60 * 🔧 chore: Update @opentelemetry/core to version 2.9.0 and clean up package-lock.json * feat: digit shortcuts select options when the popover has focus Previously a number key (1..N) only selected an option from the empty composer (handleComposerKeyDown on the textarea) — if focus moved into the popover (a row/Skip/Submit button clicked or tabbed to), the number keys went dead. Add handlePopoverKeyDown, wired to the popover container's onKeyDown so it catches digits bubbling from the focused control: a digit activates its option exactly like a click (single-select submits, multi toggles). No highlight/Enter dance on this path — the options are buttons whose action is the click, and intercepting Enter would fight the focused button. Gated on active && !locked so it no-ops while a submit is in flight. * chore: update @librechat/agents to version 3.2.61 and @opentelemetry packages to latest versions	2026-07-08 15:31:05 -04:00
Danny Avila	edc0aebdb9	🛂 fix: Re-Check execute_code Authorization on Event-Driven Tool Loads (#13912)	2026-06-23 08:30:39 -04:00
Danny Avila	49859c04a2	🗄️ fix: Gate Request-Scoped MCP Servers Out of Persistent Tool Cache (#13672) ... * 🗄️ fix: Gate Request-Scoped MCP Servers Out of Persistent Tool Cache PR #13626 established that request-scoped MCP servers (runtime OPENID/GRAPH/BODY placeholders) must not use the persistent 12h tool cache, but only gated three of five touchpoints. The panel endpoint still back-filled the cache and the OAuth callback still wrote to it, while agent loading read those entries ungated — pinning ephemeral model-spec/agent toolsets to stale definitions for up to 12h. Centralize the invariant in createMCPToolCacheService: a getServerConfig resolver dep gates both writers and a new service-owned getMCPServerTools read, so every current and future caller is covered. Callers that already hold the parsed config pass it to skip resolution; the per-call skipCache flag and duplicated call-site gates are removed in favor of the single config-based mechanism. Resolution failures fail open to preserve prior behavior. * 🩹 fix: Address Codex Review on Cache Gating - Repair getCachedTools.spec.js, which destructured the relocated getMCPServerTools directly from the module; its coverage now lives in the service-level tools.spec.ts. - Resolve the merged (Config-tier-aware) server config in the OAuth callback before writing tool definitions, so the cache gate detects request-scoped servers supplied via admin Config overlays that the base registry lookup cannot see. - Discover tools actively for request-scoped servers in the panel endpoint via ephemeral reinitialization: such servers have no stored app/user connections, so the previous getServerToolFunctions fallback returned an empty toolset once the cache read was gated. * 🧵 fix: Address Second Codex Review on Cache Gating - Resolve the merged server config before the OAuth callback reconnects, so the connection itself uses Config-tier overlays rather than only the subsequent cache write. - Pass Config-tier candidates into the panel's request-scoped discovery, matching the reinitialize route: reinitMCPServer forwards configServers (not the provided serverConfig) to its OAuth discovery fallback. - Document the accepted read-path trade-off: the gate resolver sees base configs only, all writers pass merged configs, so a pre-gating or overlay-divergent entry survives at most one cache TTL. * 🚏 chore: Rework Cache Gating for BODY-Only Request Scoping After #13673 narrowed requiresEphemeralUserConnection to BODY placeholders, the central gate follows the predicate unchanged, but the panel's active discovery no longer serves a purpose: the only remaining request-scoped class cannot connect outside a chat turn, so the reinitialization attempt would always fail at the missing-body check. Remove that path; OpenID/Graph servers are persistent user-scoped again and flow through the stored-connection and cache lookups as before. Flip test fixtures that used OPENID placeholders to denote request-scoped configs over to BODY placeholders. * 🪟 fix: Check Config Overlays in Agent-Loading Cache Reads The cache service's registry resolver sees only base YAML/DB configs, so a BODY placeholder introduced by a request-tier Config overlay was invisible to the gate on the agent-loading read path: model-spec and ephemeral-agent expansion could read a leftover persistent entry and pin stale concrete tool names instead of the mcp_all fresh-discovery path. Check the raw overlay candidate inline in loadEphemeralAgent and loadAddedAgent — a pure placeholder scan with no extra IO — and skip the cache read when the overlay makes the server request-scoped. Widen UserScopedConnectionConfig so raw (pre-inspection) configs qualify for the scoping predicates, which only check key presence. * 🧪 test: Guard Run-Scoped MCP Definition Handoff Boundaries The original ClickHouse breaker storm regressed precisely at field pass-through boundaries that unit tests of each end could not see: initializeAgent dropping mcpAvailableTools from its destructure, and the agent tool context losing it on the way into ON_TOOL_EXECUTE. Add direct guards on both hops: the loadTools result must surface on the initialized agent, and the captured toolExecuteOptions closure must forward it to loadToolsForExecution.	2026-06-13 11:26:49 -04:00
Danny Avila	139d61c437	🚐 fix: Reuse Request-Scoped MCP Connections per Run (#13673) ... * fix(mcp): reuse request-scoped connections per run * test(mcp): update connection factory defaults	2026-06-11 01:17:14 -04:00
Danny Avila	65bca95023	🎒 fix: Carry Request-Scoped MCP Tools into PTC Execution (#13669) ... * fix(mcp): preserve request-scoped tools for PTC execution * fix(mcp): preserve run-scoped tools on initialized agents	2026-06-10 23:48:04 -04:00
Danny Avila	7eafe317cc	🗝️ fix: Resolve MCP Runtime User and Request Placeholders (#13626) ... * fix: Resolve MCP Runtime User Placeholders * fix: Harden MCP Runtime Placeholder Connections * fix: Update MCP Source Tag Test Expectations * fix: Complete MCP Runtime Placeholder Reinit * fix: Harden MCP Request Scoped Runtime Configs * fix: Align MCP OAuth Tests With Domain Policy * fix: Harden MCP Runtime Resolution Edges * fix: Avoid MCP Runtime Reprocessing Pitfalls * fix: Reuse MCP Request Scoped Tool Discovery * fix: Validate MCP Body Runtime Fields * 🛡️ refactor: Harden runtime placeholder edges from review - Warn at inspection when a trusted server URL contains runtime placeholders but no domain allowlist restricts the resolved target - Document the three resolution sites that must stay in sync so the validated config always matches the connected one - Note the per-call connect cost of ephemeral GRAPH/BODY connections - Drop the no-op removeUserConnection in callTool's ephemeral cleanup; ephemeral connections are never stored, and removing the entry could orphan a still-connected cached connection after a config change * 🪪 fix: Cover oauth_headers, Graph URL gating, and request-scoped reconnects Address Codex review: - Resolve runtime placeholders in oauth_headers (processMCPEnv + Graph pre-pass) and include the field in placeholder detection, so OAuth discovery/token requests no longer send literals; consolidate the detection field lists into one helper - Defer the early domain gate when the URL still carries a Graph placeholder (resolved async later); the authoritative assertResolvedRuntimeConfigAllowed check still enforces policy - Bypass the 10s reconnect throttle for request-scoped servers, which re-fetch tool definitions on every message by design	2026-06-09 18:52:57 -04:00
Danny Avila	cb1d536874	📻 fix: Replay MCP OAuth Prompts for Coalesced Connections (#13565) ... * fix: Replay MCP OAuth URL for Joined Connections * chore: Sort MCP OAuth Imports * test: Restore MCP OAuth Registry Spies * fix: Replay pending MCP OAuth prompts * fix: Replay MCP OAuth on Stream Resume * fix: Preserve MCP OAuth Replay Context * chore: Format MCP OAuth Replay Context * test: Expect MCP OAuth Replay Expiry * fix: Render pending MCP OAuth prompts * chore: Clean MCP OAuth Replay Type Narrowing * fix: Stabilize new MCP OAuth chats * fix: Re-emit cached MCP OAuth prompts * fix: Replay pending OAuth for selected MCP tools * fix: Avoid stalling pending MCP OAuth replay * test: Clean MCP OAuth review findings * test: Restore MCP OAuth registry spy * fix: Resolve OAuth Typecheck Regressions * fix: Harden MCP OAuth replay edge cases * test: Cover MCP OAuth joined prompt expiry * test: Mark joined OAuth replay fixture * test: Use OAuth fixture for joined replay expiry * fix: Anchor resumed MCP OAuth prompts * fix: Seed resumable turn metadata before MCP init * test: Format resume metadata regression * fix: Prioritize resumable stream routes * fix: Preserve MCP OAuth resume message tree * test: Fix MCP OAuth Resume Test Types * fix: Replay MCP OAuth Regenerate Prompts * fix: Skip OAuth-only Abort Persistence * fix: Stabilize OAuth Resume Replay * fix: Target Non-Tail Regenerate Responses * fix: Scope Regenerate Step Updates * fix: Clean Up OAuth Abort State * fix: Preserve Regenerate Branch Siblings * fix: Preserve OAuth Resume Branch State * fix: Preserve OAuth Branch Resume State * chore: Sort OAuth Resume Imports * fix: Address OAuth Resume Review Findings * test: Fix Abort Fixture Typing	2026-06-07 10:45:54 -04:00
janluedemann-esome	2ed59ac98a	🔐 fix: Handle Multiple Concurrent MCP OAuth Login Prompts (#13200) ... * fix: handle multiple MCP OAuth prompts * fix: address MCP OAuth review feedback * fix: address MCP OAuth prompt lifecycle review * fix: narrow OAuth prompt slot cleanup * fix: format OAuth prompt test --------- Co-authored-by: Danny Avila <danny@librechat.ai>	2026-06-05 17:18:24 -04:00
jcbartle	268f095c1a	🔒 feat: Add On-Behalf-Of (OBO) token exchange support for MCP Servers (#13429) ... * Add OBO (On-Behalf-Of) token exchange support for MCP server connections Enables transparent authentication to Entra ID-backed MCP servers using the logged-in user's federated token via the OAuth 2.0 jwt-bearer grant. Configured via obo.scopes in librechat.yaml server config. - Extract generic OboTokenService from GraphTokenService (jwt-bearer grant + cache) - Refactor GraphTokenService to thin wrapper delegating to OboTokenService - Add obo schema field to BaseOptionsSchema in data-provider - Add resolveOboToken in packages/api/src/mcp/oauth/obo.ts (validates federated token, calls resolver, returns MCPOAuthTokens) - Wire oboTokenResolver through MCPConnectionFactory, MCPManager, UserConnectionManager - OBO tokens injected via request headers (not OAuth transport), refreshed on each tool call - Explicit error on OBO failure (no fallthrough to standard OAuth redirect) - Add unit tests for both resolveOboToken (9 tests) and exchangeOboToken (14 tests) * Add OBO authentication option to MCP server UI configuration Enable users to configure On-Behalf-Of (OBO) token exchange for MCP servers created via the UI (MongoDB-stored), in addition to the existing YAML-based configuration. - Add "On-Behalf-Of (OBO)" radio option to MCP server auth section with scopes input field - Remove obo from omitServerManagedFields so the field passes UI schema validation - Add OBO to AuthTypeEnum, obo_scopes to AuthConfig, and OBO handling in form defaults and submission - Add .min(1) validation on obo.scopes to reject empty strings - Add English localization keys: com_ui_obo, com_ui_obo_scopes, com_ui_obo_scopes_description - Add 5 schema validation tests for OBO field acceptance, transport compatibility, and edge cases * 🧊 fix: Add obo to safe properties in redactServerSecrets. Fixes the OBO configuration not showing up in the MCP UI after app restart * Address linter errors * 🧊 fix: fail closed on OBO refresh errors and retry transient token exchange failures - stop tool calls from falling back to stale Authorization headers when per-call OBO refresh fails - add one-time retry for transient Entra OBO exchange failures (network/429/5xx) - preserve structured OBO failure reasons and retryability in resolveOboToken - improve OBO auth error messaging for connection setup and tool execution - add tests for transient vs permanent OBO failure paths * Addressing linting errors / warnings * 🧊 fix: isolate OBO MCP auth to user-scoped connections - block OBO-enabled servers from app-level shared MCP connections - bypass shared connection lookup for OBO servers in MCPManager.getConnection - add regressions covering OBO connection scoping and preserve non-OBO app connection reuse * 🛠️ refactor: centralize MCP user-scoped connection policy - add shared requiresUserScopedConnection helper for OAuth, OBO, and customUserVars - use the shared predicate in MCPManager and ConnectionsRepository - add utils coverage for user-scoped connection policy * 🧊 fix: restrict MCP OBO config to header-capable transports - Move OBO configuration out of the shared MCP base options schema and allow it only on SSE and streamable-http transports, where request headers are applied. - Explicitly reject OBO on stdio and websocket configs to avoid accepted-but- nonfunctional server definitions. Add schema coverage for admin/config parsing and user-input websocket validation. * 🧊 fix: single-flight concurrent OBO token exchanges Concurrent tool calls that arrive on a cache miss were each issuing their own jwt-bearer request to the IdP. Under that fan-out, Entra intermittently returned errors that the retry classifier saw as non-retryable, surfacing as: "The identity provider rejected the OBO token exchange. Cannot execute tool <name>. Re-authenticate the user or verify the configured OBO scopes and retry." A user retry then hit the populated cache and succeeded, which matches the observed flakiness — the cache was empty at the moment of fan-out but populated by the time the user clicked retry. - Coalesce concurrent exchanges in `OboTokenService.exchangeOboToken` keyed by `${openidId}:${scopes}`. Callers that arrive while an exchange is in flight share the same upstream request and receive the same result. `fromCache=false` continues to force a fresh, independent exchange (and is not joined by `fromCache=true` callers). The IdP call, single-retry path, and cache write are unchanged — they were moved into a `performOboExchange` helper so the coalescing wrapper stays small. - Tests cover: coalescing on the same key, isolation between different keys, cleanup on success, cleanup on failure, and the `fromCache=false` bypass. * 🔒 feat: gate MCP OBO config behind MCP_SERVERS.CONFIGURE_OBO permission OBO silently mints per-user delegated tokens from the caller's federated access token and forwards them to whatever URL the server config points at. Previously, anyone with MCP_SERVERS.CREATE could configure obo.scopes — so if server creation is ever delegated beyond admins, a user could stand up an attacker-controlled server, attach it to a shared agent, and exfiltrate other users' downstream tokens on tool invocation. Add a dedicated MCP_SERVERS.CONFIGURE_OBO permission (ADMIN: true, USER: false by default) and enforce it at three layers so the safety property no longer depends on CREATE staying admin-only: - Create/update: POST/PATCH /api/mcp/servers returns 403 when the body carries `obo` and the caller's role lacks the permission. - Runtime fail-closed: for DB-sourced configs, MCPConnectionFactory and MCPManager.callTool re-check the original author's role before each OBO exchange. If the author has been downgraded, the exchange is skipped (factory) or refused (callTool) — retained configs lose their privileges automatically. - UI: the OBO option is hidden in the MCP server dialog for users without the permission; a CONFIGURE_OBO toggle is exposed in the MCP admin role editor. Existing role docs receive the new sub-key via the permission backfill in updateInterfacePermissions on next startup, preserving any operator-set values. YAML/Config-sourced server configs are unaffected since they're admin-controlled at the deployment level. * 🧊 fix: wire OBO machinery for servers with requiresOAuth: false The discovery and user-connection paths gated OAuth wiring (flow manager, token methods, oboTokenResolver, oboTrustChecker) behind isOAuthServer(), which only considers requiresOAuth/oauth fields. A DB-stored OBO server with requiresOAuth: false therefore landed in the non-OAuth branch, never received an oboTokenResolver, and the factory's usesObo getter evaluated to false — sending a bare request that the upstream rejected with invalid_token. Add requiresOAuthMachinery() (OAuth OR OBO) and use it at those two gates. isOAuthServer remains for the OAuth-handshake-only check (shouldInitiateOAuthBeforeConnect), where OBO must not initiate a handshake. Plumb the OBO resolver/trust-checker through ToolDiscoveryOptions so reinitMCPServer can pass them on the discovery path. * 🧊 fix: lock all OBO-target fields (URL, proxy, headers, auth) without CONFIGURE_OBO The CONFIGURE_OBO permission was meant to gate control of the endpoint that receives OBO-minted per-user delegated tokens and the scopes that are requested. The previous frontend lock + backend gate only covered obo.scopes and the auth section, leaving url/proxy/headers/etc. editable by anyone with UPDATE — meaning a non-permission user could still redirect an existing OBO server's token flow to an attacker endpoint. Switch to an allowlist policy: when editing an OBO server without CONFIGURE_OBO, only title/description/iconPath are mutable. Backend rejects any other field change with 403; frontend disables the non-allowlist sections (URL, transport, auth, trust) via fieldset. The comparison surface (MCP_USER_INPUT_FIELDS) is derived from MCPServerUserInputSchema's union members so it stays in sync with the schema. New schema fields land in the locked set by default — adding to the allowlist is the only way to unlock them, which preserves the security-review boundary. * 🧊 fix: skip unauthenticated MCP inspection for OBO-only servers MCPServerInspector.inspectServer() ran an unauthenticated temp connection unless the config had requiresOAuth or customUserVars set. For OBO-only servers without standard MCP OAuth advertisement, this caused MCPConnectionFactory.create to attempt the connection without a user or oboTokenResolver — failing on servers that reject the MCP initialize handshake without a valid bearer token, which surfaced as MCP_INSPECTION_FAILED on create/update. Add `obo` to the skip list alongside requiresOAuth and customUserVars, matching the existing pattern for user-scoped auth modes. * Addressed linting error: watchedTitle is declared but never referenced (the auto-fill logic at line 156 uses getValues('title') instead). Deleted constant.	2026-06-01 22:36:18 -04:00
Danny Avila	479e9d59b7	🧠 refactor: Memoize MCP Permission Checks Per Request (#13419)	2026-05-30 18:32:06 -04:00
Danny Avila	100871c3ec	🛂 fix: Enforce MCP Permissions for Agent Tools (#13174) ... * fix: Enforce MCP Permissions for Agent Tools * fix: Measure MCP Image Limit by Decoded Size * fix: gate cached MCP tools and tighten remote image URL detection Addresses Codex review findings on the MCP permissions PR: - filterAuthorizedTools previously fast-accepted any tool present in the global tool cache before reaching the MCP-use permission gate. App-level MCP tools (keyed `name_mcp_server` by MCPServerInspector and merged into the cache via mergeAppTools) therefore bypassed the canUseMCP check, letting a user without MCP_SERVERS.USE persist/bind them. Route all MCP-delimited tools through the permission + server-access gate regardless of cache presence. - assertImageDataWithinLimit / image formatter used startsWith("http") to skip the size cap, which also matched base64 payloads that happen to begin with those chars. Require http:// or https:// via a shared isRemoteImageUrl helper so oversized inline base64 can no longer bypass MCP_IMAGE_DATA_MAX_BYTES. Adds regression tests for both paths. * fix: address Codex round-2 findings on MCP permissions PR - parsers.ts: parseAsString dropped the image payload for unrecognized providers, returning only `Image result: <mimeType>`. Pre-PR these items survived via JSON.stringify(item). Keep the size guard but fall through to JSON.stringify so the data/URL is preserved. - MCP.js: the runtime MCP-use check only read `configurable.user`, so paths that propagate `user_id` only (e.g. the OpenAI-compatible API in agents/openai/service.ts) rejected every MCP tool call for an authenticated user. Add resolveMCPPermissionUser: use the safe user directly when it already carries a role (no extra DB call), otherwise fall back to loading the role by user_id. Update fail-closed tests to the resolved behavior. - v1.js: the update path only re-filtered newly added MCP tools, so a user who lost MCP_SERVERS.USE kept existing MCP bindings on edit while create/duplicate/revert stripped them. Strip all MCP tools on update when the permission is revoked; keep the narrower new-tool gating (and disconnect/registry preservation) when it is intact. Updates and adds regression tests for all three paths. * fix: populate safe user at producer instead of resolving in runtime MCP check Corrects the Finding B approach from the previous commit. Rather than loading the user by id inside the runtime MCP permission check, populate `configurable.user` (and createRun's `user`) with the full safe user at the producer, matching the in-repo agent controllers (responses.js / openai.js) which already pass `createSafeUser(req.user)`. - service.ts: derive `safeUser` via createSafeUser(req.user) and pass it to both createRun and processStream's configurable, so the role-bearing identity reaches the runtime `userCanUseMCPServers(configurable.user)` check. Falls back to a bare id when the host app attached no user, which correctly leaves MCP gated (fail closed). - MCP.js: revert the resolveMCPPermissionUser DB-load fallback; the runtime check again reads configurable.user directly and fails closed when absent (defense in depth). - MCP.spec.js: revert to the matching runtime test expectations. * test: cover safe-user propagation in createAgentChatCompletion Adds a focused spec for the OpenAI-compatible chat completion service (the producer fixed for Codex Finding B). Injects mocked deps and asserts that createRun and processStream's configurable.user carry the role from req.user (with sensitive fields stripped by createSafeUser), and that an unauthenticated request falls back to a bare { id: 'api-user' } so the runtime MCP check fails closed. * fix: address Codex round-3 findings + TS6133 - MCP.js (P1): the assistants required-action path invokes tool._call( toolInput) with no LangChain config, so the runtime check saw no configurable.user and rejected authorized users. createToolInstance now captures the creation-time user (req.user via createMCPTool) and _call falls back to it for both the permission check and userId. Still fails closed when neither config nor captured user carries a role. - v1.js (P2): the update-path isMCPTool used a bare mcp_delimiter substring check, misclassifying action tools whose operationId contains "_mcp_" (e.g. sync_mcp_state_action_...) as MCP and dropping them on a permission-revoked edit. Delegate to the canonical isActionTool so only real MCP tools are gated. Regression test added. - service.ts: drop the now-unused IUser import (TS6133); derive reqUser's type from createSafeUser's own parameter instead. * fix: resolve TS7022 self-reference in service.spec mock res The mock response object referenced `res` inside its own `status`/`json` initializers without a type annotation, so tsc inferred `res` as `any` (TS7022). Annotate the object and assign the self-referencing chainable methods after declaration. * fix: correct round-4 findings (isActionTool import, captured user, partial-update) - v1.js: import isActionTool from librechat-data-provider (its real export; @librechat/api does not export it, so the prior import was undefined and threw TypeError). Exclude action tools from MCP classification in both the main filterAuthorizedTools loop and the update path, so action tools whose operationId contains _mcp_ (e.g. sync_mcp_state_action_...) are preserved regardless of MCP permission. - v1.js: evaluate the effective tool set (updateData.tools ?? existingAgent.tools) so a tools-less PATCH by a user who lost MCP_SERVERS.USE still strips stale MCP bindings, matching create/duplicate/revert. - MCP.js: createToolInstance now receives the construction-time user and _call falls back to it (permissionUser) when configurable.user is absent, fixing the assistants required-action path that invokes _call without a config and resolving the capturedUser no-undef/ReferenceError. - Tests: action-tool preservation (authorized + denied), tools-less revocation PATCH, updated revocation test to expect all MCP tools stripped. Affected specs pass locally: MCP 49/49, filterAuthorizedTools 49/49. * fix: guard isActionTool against non-string tools; correct actionDelimiter import Two test regressions from the prior commit: - The main filterAuthorizedTools loop called isActionTool(tool) directly, but isActionTool does toolName.indexOf(...) and throws on null/undefined. Compute isActionToolName = typeof tool === 'string' && isActionTool(tool) once and reuse it, restoring graceful null/undefined handling. - The action-tool test referenced Constants.actionDelimiter (undefined); actionDelimiter is a standalone librechat-data-provider export. Import and use it directly. filterAuthorizedTools 36/36 and MCP 40/40 pass locally. * fix: address MCP permission review follow-ups * fix: preserve shared agent MCP tools	2026-05-30 16:19:49 -04:00
Max	1746153c17	🪬 fix: Skip MCP Tools When Required Custom User Vars Are Unset (#13152) ... * fix: skip MCP tools when required customUserVars are unset (#10969) * fix: whitespace-only values Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com> * fix: guard MCP registry lookup and unknown server config in customUserVars gate * fix: fail closed on MCP registry lookup errors --------- Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com> Co-authored-by: Danny Avila <danny@librechat.ai>	2026-05-23 09:24:29 -04:00
Danny Avila	bd64251eb9	🪪 fix: Prevent MCP Server Name Collisions (#13256) ... * fix: prevent MCP server name collisions * chore: address MCP registry review nits * fix: reserve MCP config names from request context * chore: format MCP registry changes * chore: address MCP collision review findings	2026-05-22 20:46:14 -04:00
Danny Avila	749eb06e67	🧭 fix: Reduce MCP Registry ACL Lookups (#13195)	2026-05-19 17:16:37 -04:00
Danny Avila	5bab22d236	🛡️ fix: Gate Bash PTC Capabilities (#13053)	2026-05-10 21:23:02 -04:00
Danny Avila	715a4a5fc1	🧰 refactor: Use Bash PTC for Agent Tools (#13042) ... * fix: Use Bash PTC for programmatic agent tools * fix: Preserve legacy PTC event calls	2026-05-09 16:31:09 -04:00
Danny Avila	f839a447e1	🧬 fix: Subagent MCP requestBody Propagation (bump @librechat/agents to 3.1.78 + cleanup) (#12959) ... * 📦 chore: bump `@librechat/agents` to v3.1.78 v3.1.78 ships [danny-avila/agents#147](https://github.com/danny-avila/agents/pull/147), which makes `SubagentExecutor` inherit the parent invocation's `configurable` (with `thread_id`/`run_id`/`parent_run_id` scrubbed) into the child workflow. Subagent tool dispatches through the parent's `ON_TOOL_EXECUTE` handler now arrive with parent's `requestBody`, `user`, `userMCPAuthMap`, etc. — so `{{LIBRECHAT_BODY_*}}` placeholder substitution and per-user MCP connection lookup work for subagent tool calls the same way they do for the parent agent. Note: `package-lock.json` will need an `npm install` refresh once v3.1.78 lands on the registry. The user/user_id injection added in PR #12950 stays as defense-in-depth. * 🗑️ refactor: drop redundant user/user_id injection from `loadToolsForExecution` `@librechat/agents@3.1.78` (via danny-avila/agents#147) makes `SubagentExecutor` forward the parent's `configurable` verbatim into the child workflow. Subagent `ON_TOOL_EXECUTE` dispatches now arrive with parent's `user` / `user_id` already in `data.configurable` — making the host-side injection added in #12950 a no-op. Removes: - The conditional `user: createSafeUser(req.user); user_id: req.user.id` block in `loadToolsForExecution` (req.user.id-guarded so the `'api-user'` fallback in Responses/OpenAI controllers is preserved). - The unused `createSafeUser` import. - The 4 unit tests covering the now-deleted behavior. The merge in `handlers.ts` (`{ ...configurable, ...toolConfigurable }`) still produces a `mergedConfigurable` with the right user identity for both parent and subagent paths — the values just come from `configurable` (forwarded by the SDK) rather than `toolConfigurable`. Other fixes from #12950 stay (IUser.id narrowing, the env.ts / google/initialize.ts / remoteAgentAuth.ts TS-warning fixes) — they were independent of the subagent identity propagation issue. * 📦 chore: update `@librechat/agents` to v3.1.78 This update reflects the transition from the development version `3.1.78-dev.0` to the stable release `3.1.78`. The package-lock.json has been refreshed to ensure consistency with the new version, including updated integrity checks and resolved URLs for the package. This change is part of ongoing improvements to enhance the functionality and stability of the agents module.	2026-05-05 22:07:26 -04:00
Danny Avila	bff9bfea87	🐛 fix: Propagate User Identity to Subagent MCP Tool Calls (#12950) ... * 🐛 fix: Propagate User Identity to Subagent MCP Tool Calls The `@librechat/agents` SDK's `SubagentExecutor` invokes the child workflow with a fresh configurable of `{ thread_id }` only — the parent's `user` / `user_id` are dropped on the way into the child graph. The child's `ToolNode` then dispatches `ON_TOOL_EXECUTE` to the parent's handler, which merges `{ ...configurable, ...toolConfigurable }`, but neither side carries user identity for subagents. Downstream MCP tools read `config.configurable.user?.id || user_id` and got `undefined`, so `MCPManager.getConnection` fell through to the "No connection found for server X" error path — it can't reach the user-connection lookup without a userId. Re-inject `user` (via `createSafeUser`) and `user_id` from `req.user` into the configurable returned by `loadToolsForExecution`. This is the single point all controllers (chat, Responses API, OpenAI-compat) flow through. For the parent agent it's a no-op (outer config already carries the same values); for subagents it fills the gap so MCP connection lookup, user-placeholder substitution, and tools that read configurable.user all work correctly. * 🐛 fix: Preserve `api-user` Fallback When Injecting Subagent Identity Codex review pointed out that the prior commit unconditionally wrote `user_id: req.user?.id` (and `user`) into `toolConfigurable`. The handler merges via `{ ...configurable, ...toolConfigurable }` — `toolConfigurable` wins — so when `req.user` is absent, this overwrote the outer config's `'api-user'` fallback (set by `responses.js` / `openai.js` for the unauthenticated API-key path) with `undefined`, breaking MCP connection lookup for that path. Only inject the keys when `req.user.id` is truthy. Omitting them lets the merge preserve whatever the outer configurable already had. Tests updated to assert key omission for `req.user` undefined / null / present without `id`. * 🩹 fix: Narrow `IUser.id` to required string `IUser` extends mongoose `Document`, which types `id?: any` (the optional virtual). At runtime `id` is always `_id.toString()` for a hydrated doc, so narrow the type to a required string. Closes two `@rollup/plugin-typescript` TS2322 warnings introduced by PR #12450 (OIDC Bearer Token Authentication for Remote Agent API) where `req.user = userResolution.user` and the `(req: Request, res: Response, next: NextFunction)` signature both failed against the project's local `Express.User` augmentation (`{ [key: string]: any; id: string; }`) because `IUser.id` was `any`/optional. Narrowing here fixes both at the source rather than casting at every assignment site. * 🩹 fix: Resolve TS Build Warnings Surfaced by `IUser.id` Narrowing Three rollup TS plugin warnings surfaced after narrowing `IUser.id` from `any` to `string`: - `utils/env.ts:95` — `safeUser[field] = user[field]` failed strict checking because indexed write through a union-typed key collapses the LHS to the intersection of all field write types (i.e., `undefined` when fields have mixed types). The previous `id?: any` on IUser had been masking this. Switch to `Object.assign(safeUser, { [field]: user[field] })` which widens the assignment. - `endpoints/google/initialize.ts:35` — `getUserKey({ userId: req.user?.id, ... })` failed because `req.user?.id` is now `string | undefined` (no longer `any`). Match the pattern already used in `endpoints/openAI/initialize.ts:49`: `req.user?.id ?? ''`. - `middleware/remoteAgentAuth.ts:465` — pre-existing, unrelated to the IUser change. The local (gitignored) `express.d.ts` augments `express.Request` but not `express-serve-static-core.Request`, so the explicit `(req: Request, ...)` annotation imported from `'express'` resolves to a Request whose `req.user` differs from the one `RequestHandler` expects internally. Type the closure as `RequestHandler` directly so TS infers params from the augmented type. * 🩹 fix: Cast `RemoteAgentAuth` Closure to `RequestHandler` My previous attempt removed the explicit `req: Request` annotation on the closure to side-step the outer `RequestHandler` mismatch. That shifted the error to every helper call site inside the closure (`getConfigOptions(req)`, `runApiKeyAuth(req, ...)` at 467/474/493/ 512/531), because the helpers annotate their params with `express.Request` (which has the local `Request.user` augmentation), while the unannotated closure inferred `req` as `express-serve-static-core.Request` (no augmentation). Reproduced locally by stubbing the gitignored `src/types/express.d.ts`. Right approach: keep the explicit `req: Request` annotation so the closure body matches the helpers' types, then cast at the return — `RequestHandler`'s internal `Request` resolves through `express-serve-static-core` and lacks the augmentation, so the cast is the boundary that bridges the two views of `req.user`. Verified against a build with the local express.d.ts stub: zero warnings on `remoteAgentAuth.ts`, `env.ts`, and `google/initialize.ts`.	2026-05-05 12:19:50 +09:00
Caesar Isidro Va-ay	9b9a86d17d	🔀 fix: Resolve Action Tools by Exact Name to Prevent Multi-Action Domain Collision (#12594) ... * 🐛 fix: resolve Action tools by exact tool name to prevent multi-action collision When two OpenAPI Actions on the same Agent share a hostname, the second action's entry overwrote the first in the encoded-domain Map and one action's tools silently disappeared from the LLM payload. The buggy resolution loop also used substring matching, which caused similar shadowing for any encoded-domain prefix overlap. This change builds a Map keyed on the full tool name (`<operationId>_action_<encoded-domain>`) directly, mirroring the exact lookup pattern that getActionToolDefinitions already uses. Each function in an action's spec gets its own slot, so two actions sharing a hostname no longer collide. Both the new and the legacy domain encodings are registered for each function so agents whose stored tool names predate the current encoding still resolve. Applied at all three call sites that had the buggy pattern: - processRequiredActions (assistants/threads path) - loadAgentTools (agent build path) - loadActionToolsForExecution (agent execution path) Adds three regression tests covering both ordering directions and the execution path. Tests fail without the fix and pass with it. * 🐛 fix: Normalize action tool name at lookup + cover assistants path Follow-up to the multi-action domain collision fix. Addresses PR #12594 review feedback: **Must-fix #1 — short-hostname lookup mismatch.** The toolToAction map is keyed on the `_`-collapsed domain, but `agent.tools` and `currentAction.tool` persist the raw `domainParser(..., true)` output, which for hostnames ≤ ENCODED_DOMAIN_LENGTH is a `---`-separated string (e.g. `medium---com`). Exact-match `Map.get()` missed those keys and silently dropped the tool. Fix: normalize every incoming tool name through a new `normalizeActionToolName` helper before the lookup in `loadAgentTools`, `processRequiredActions`, and `loadActionToolsForExecution`. **Must-fix #2 — assistants path coverage.** `processRequiredActions` received the same structural rewrite but had zero tests. Added a regression test under `multi-action domain collision regression` that drives two shared-hostname actions through the assistants path and asserts each tool reaches its own request builder. **Must-fix #3 — legacy encoding branch coverage.** The `if (legacyNormalized !== normalizedDomain)` registration was never exercised by any test. Added a test where `agent.tools` stores the legacy-format name and asserts it still resolves. **Should-fix #4 — DRY the registration loop.** Extracted `registerActionTools({ toolToAction, functionSignatures, normalizedDomain, legacyNormalized, makeEntry })`. All three call sites now share the same key-building logic; the key template lives in one place. **Should-fix #5 — remove stale optional chaining.** In `loadActionToolsForExecution`, `functionSignature?.description ?? ''` became `functionSignature.description` — `sig` is always defined by the iterator, matching the style of `loadAgentTools`. **Should-fix #6 — drop unreachable `!requestBuilder` guard.** Entries in `processRequiredActions` are now pre-built with `requestBuilder: requestBuilders[sig.name]`, which `openapiToFunction` always produces alongside the signature, so the guard is dead. **Should-fix #7 — unwrap `actionSetsData`.** It now holds a bare `Map` instead of `{ toolToAction }`; the sentinel `!actionSetsData` check still works because `new Map()` is truthy. Also added a short-hostname regression test (`loadAgentTools resolves raw ---separated tool names`) that reproduces Must-fix #1: it fails against the previous commit (0 create calls) and passes with the normalization in place. 41 tests, all passing. The 3 new regression tests are under `multi-action domain collision regression` and cover the assistants path, the legacy encoding branch, and the short-hostname lookup path. * 🐛 fix: Tighten registerActionTools key handling and assistants test Follow-up to d643444 addressing the second review pass on PR #12594. **ESLint** — Two prettier errors in the spec file (multi-line arrow function bodies that should fit on one line). Auto-fixed. **[MINOR] operationId containing `---` → key mismatch.** The lookup path collapsed every `actionDomainSeparator` sequence in the full tool name, but the registration path passed `sig.name` through unchanged. A `---` that survived into an operationId would shift the underscore boundary at lookup and miss its own key. Fix in `registerActionTools`: normalize `sig.name` with the same helper so registration and lookup always agree on the canonical form. `sanitizeOperationId` strips the characters that produce `---` in practice, so this is theoretical hardening, not a fix for a known reproducer. **[MINOR] Same-operationId + same-hostname silent overwrite.** Two actions sharing both an operationId and a hostname still produced a silent `Map.set()` overwrite (the new key is identical, so neither the operationId nor the domain disambiguates). Added a `setKey` helper inside `registerActionTools` that logs a `[Actions] operationId collision: ...` warning whenever a key is already present, naming the overwriting action_id. The silent-overwrite mode from the original bug cannot reappear under a different disguise without surfacing in the logs. **[NIT] processRequiredActions test simulated a runtime crash.** `mockCreateActionTool` returned a tool with `_call: jest.fn()`, which resolves to `undefined`. `processRequiredActions` chains `.then(handleToolOutput).catch(handleToolError)` directly onto that return, so `undefined.then(...)` threw synchronously and the outer try/catch funneled the error into `handleToolError`. Creation count assertions still passed because `createActionTool` runs before the crash, but the test was silently exercising the failure path. Updated the global mock to `_call: jest.fn().mockResolvedValue('{"status":"ok"}')` so the success path runs end-to-end. The assistants regression test now executes in ~5ms instead of ~90ms, which corroborates that it's no longer hitting the synchronous throw. **[NIT] Duplicated rationale comments.** All three call sites carried multi-line comment blocks restating why we key on the full tool name. That rationale now lives canonically in `registerActionTools`'s JSDoc; the inline blocks collapsed to `// See registerActionTools for the key-shape rationale.` Net -22 lines of comments. 41/41 tests still pass; lint is clean. * 🐛 fix: Scope tool-name normalization to the encoded-domain suffix Follow-up to f22228e addressing the Codex P1 on PR #12594. **Regression.** The previous commit normalized the entire tool name (`normalizeActionToolName(sig.name)` at registration, full-name `.replace()` at lookup) to handle operationIds that theoretically contained `---`. But `openapiToFunction` uses user-supplied operationIds verbatim and the fallback `sanitizeOperationId` only strips characters outside `[a-zA-Z0-9_-]`, so specs can legitimately produce operationIds like `get_foo---bar` and `get_foo_bar` side by side. Collapsing `---` to `_` across the entire key merged those two into a single map slot — one silently overwrote the other, and both tool requests routed to the surviving entry's request builder. **Fix.** Limit normalization to the encoded-domain portion of the full tool name, i.e. the substring after the last `actionDelimiter`. The operationId half is left verbatim, so hyphens-vs-underscores remain disambiguating. The short-hostname bug (Must-fix #1 from the original review) is still covered because the `---` → `_` collapse still happens on the domain suffix where it matters: ```js const normalizeActionToolName = (toolName) => { const delimiterIndex = toolName.lastIndexOf(actionDelimiter); if (delimiterIndex === -1) return toolName; const prefixEnd = delimiterIndex + actionDelimiter.length; const encodedDomain = toolName.slice(prefixEnd); return toolName.slice(0, prefixEnd) + encodedDomain.replace(domainSeparatorRegex, '_'); }; ``` `registerActionTools` reverts to `sig.name` verbatim — no more `normalizeActionToolName(sig.name)` ahead of the key build. **Regression test.** Added `loadAgentTools distinguishes operationIds that differ only by ---` vs `_`` under `multi-action domain collision regression`. It loads two actions sharing a hostname whose operationIds are `get_foo---bar` and `get_foo_bar` respectively, each pointing at a different path (`/foo-bar`, `/foo_bar`), and asserts that each tool resolves to its own request builder. Verified the test fails against f22228e (`hyphenTool` resolves to `/foo_bar` — the sibling's builder) and passes with this commit. 42/42 tests pass; lint clean.	2026-04-13 09:08:06 -04:00
Danny Avila	275af48592	🎯 fix: MCP Tool Misclassification from Action Delimiter Collision (#12512) ... * fix: prevent MCP tools with `_action` in name from being misclassified as OpenAPI action tools Add `isActionTool()` helper that checks for the `_action_` delimiter while guarding against cross-delimiter collision with `_mcp_`. Replace all `includes(actionDelimiter)` classification checks with the new helper across backend and frontend. * test: add coverage for MCP/action cross-delimiter collision Verify that `isActionTool` correctly rejects MCP tool names containing `_action` and that `loadAgentTools` does not filter them based on `actionsEnabled`. Add ToolIcon and definitions test cases. * fix: simplify isActionTool to handle all MCP name patterns - Use `!toolName.includes('_mcp_')` instead of checking only after the first `_action_` occurrence, which missed MCP tools with `_action_` in the middle of their name (e.g. `get_action_data_mcp_myserver`). - Reference `Constants.mcp_delimiter` value via a local const to avoid circular import from config.ts, with a comment explaining why. - Remove dead `actionDelimiter` import from definitions.ts. - Replace double-filter with single-pass partition in loadToolsForExecution. - Add test for mid-name `_action_` collision case. * fix: narrow MCP exclusion to delimiter position in isActionTool Only reject when `_mcp_` appears after `_action_` (the MCP suffix position). `_mcp_` before `_action_` is part of the operationId and is valid — e.g. `sync_mcp_state_action_api---example---com` is a legitimate action tool whose operationId happens to contain `_mcp_`. * fix: document positional _mcp_ guard and known RFC-invalid domain limitation Expand JSDoc on isActionTool to explain the action/MCP format disambiguation and the theoretical false negative for non-RFC-compliant domains containing `_mcp_`. Add test documenting this known edge case.	2026-04-01 22:36:21 -04:00
Danny Avila	935288f841	🏗️ feat: 3-Tier MCP Server Architecture with Config-Source Lazy Init (#12435) ... * feat: add MCPServerSource type, tenantMcpPolicy schema, and source-based dbSourced wiring - Add `tenantMcpPolicy` to `mcpSettings` in YAML config schema with `enabled`, `maxServersPerTenant`, `allowedTransports`, and `allowedDomains` - Add `MCPServerSource` type ('yaml' | 'config' | 'user') and `source` field to `ParsedServerConfig` - Change `dbSourced` determination from `!!config.dbId` to `config.source === 'user'` across MCPManager, ConnectionsRepository, UserConnectionManager, and MCPServerInspector - Set `source: 'user'` on all DB-sourced servers in ServerConfigsDB * feat: three-layer MCPServersRegistry with config cache and lazy init - Add `configCacheRepo` as third repository layer between YAML cache and DB for admin-defined config-source MCP servers - Implement `ensureConfigServers()` that identifies config-override servers from resolved `getAppConfig()` mcpConfig, lazily inspects them, and caches parsed configs with `source: 'config'` - Add `lazyInitConfigServer()` with timeout, stub-on-failure, and concurrent-init deduplication via `pendingConfigInits` map - Extend `getAllServerConfigs()` with optional `configServers` param for three-way merge: YAML → Config → User - Add `getServerConfig()` lookup through config cache layer - Add `invalidateConfigCache()` for clearing config-source inspection results on admin config mutations - Tag `source: 'yaml'` on CACHE-stored servers and `source: 'user'` on DB-stored servers in `addServer()` and `addServerStub()` * feat: wire tenant context into MCP controllers, services, and cache invalidation - Resolve config-source servers via `getAppConfig({ role, tenantId })` in `getMCPTools()` and `getMCPServersList()` controllers - Pass `ensureConfigServers()` results through `getAllServerConfigs()` for three-way merge of YAML + Config + User servers - Add tenant/role context to `getMCPSetupData()` and connection status routes via `getTenantId()` from ALS - Add `clearMcpConfigCache()` to `invalidateConfigCaches()` so admin config mutations trigger re-inspection of config-source MCP servers * feat: enforce tenantMcpPolicy on admin config mcpServers mutations - Add `validateMcpServerPolicy()` helper that checks mcpServers against operator-defined `tenantMcpPolicy` (enabled, maxServersPerTenant, allowedTransports, allowedDomains) - Wire validation into `upsertConfigOverrides` and `patchConfigField` handlers — rejects with 403 when policy is violated - Infer transport type from config shape (command → stdio, url protocol → websocket/sse, type field → streamable-http) - Validate server domains against policy allowlist when configured * revert: remove tenantMcpPolicy schema and enforcement The existing admin config CRUD routes already provide the mechanism for granular MCP server prepopulation (groups, roles, users). The tenantMcpPolicy gating adds unnecessary complexity that can be revisited if needed in the future. - Remove tenantMcpPolicy from mcpSettings Zod schema - Remove validateMcpServerPolicy helper and TenantMcpPolicy interface - Remove policy enforcement from upsertConfigOverrides and patchConfigField handlers * test: update test assertions for source field and config-server wiring - Use objectContaining in MCPServersRegistry reset test to account for new source: 'yaml' field on CACHE-stored configs - Add getTenantId and ensureConfigServers mocks to MCP route tests - Add getAppConfig mock to route test Config service mock - Update getMCPSetupData assertion to expect second options argument - Update getAllServerConfigs assertions for new configServers parameter * fix: disconnect active connections when config-source servers are evicted When admin config overrides change and config-source MCP servers are removed, the invalidation now proactively disconnects active connections for evicted servers instead of leaving them lingering until timeout. - Return evicted server names from invalidateConfigCache() - Disconnect app-level connections for evicted servers in clearMcpConfigCache() via MCPManager.appConnections.disconnect() * fix: address code review findings (CRITICAL, MAJOR, MINOR) CRITICAL fixes: - Scope configCacheRepo keys by config content hash to prevent cross-tenant cache poisoning when two tenants define the same server name with different configurations - Change dbSourced checks from `source === 'user'` to `source !== 'yaml' && source !== 'config'` so undefined source (pre-upgrade cached configs) fails closed to restricted mode MAJOR fixes: - Derive OAuth servers from already-computed mcpConfig instead of calling getOAuthServers() separately — config-source OAuth servers are now properly detected - Add parseInt radix (10) and NaN guard with fallback to 30_000 for CONFIG_SERVER_INIT_TIMEOUT_MS - Add CONFIG_CACHE_NAMESPACE to aggregate-key branch in ServerConfigsCacheFactory to avoid SCAN-based Redis stalls - Remove `if (role || tenantId)` guard in getMCPSetupData — config servers now always resolve regardless of tenant context MINOR fixes: - Extract resolveAllMcpConfigs() helper in mcp controller to eliminate 3x copy-pasted config resolution boilerplate - Distinguish "not initialized" from real errors in clearMcpConfigCache — log actual failures instead of swallowing - Remove narrative inline comments per style guide - Remove dead try/catch inside Promise.allSettled in ensureConfigServers (inner method never throws) - Memoize YAML server names to avoid repeated cacheConfigsRepo.getAll() calls per request Test updates: - Add ensureConfigServers mock to registry test fixtures - Update getMCPSetupData assertions for inline OAuth derivation * fix: address code review findings (CRITICAL, MAJOR, MINOR) CRITICAL fixes: - Break circular dependency: move CONFIG_CACHE_NAMESPACE from MCPServersRegistry to ServerConfigsCacheFactory - Fix dbSourced fail-closed: use source field when present, fall back to legacy dbId check when absent (backward-compatible with pre-upgrade cached configs that lack source field) MAJOR fixes: - Add CONFIG_CACHE_NAMESPACE to aggregate-key set in ServerConfigsCacheFactory to avoid SCAN-based Redis stalls - Add comprehensive test suite (ensureConfigServers.test.ts, 18 tests) covering lazy init, stub-on-failure, cross-tenant isolation via config hash keys, concurrent deduplication, merge order, and cache invalidation MINOR fixes: - Update MCPServerInspector test assertion for dbSourced change * fix: restore getServerConfig lookup for config-source servers (NEW-1) Add configNameToKey map that indexes server name → hash-based cache key for O(1) lookup by name in getServerConfig. This restores the config cache layer that was dropped when hash-based keys were introduced. Without this fix, config-source servers appeared in tool listings (via getAllServerConfigs) but getServerConfig returned undefined, breaking all connection and tool call paths. - Populate configNameToKey in ensureSingleConfigServer - Clear configNameToKey in invalidateConfigCache and reset - Clear stale read-through cache entries after lazy init - Remove dead code in invalidateConfigCache (config.title, key parsing) - Add getServerConfig tests for config-source server lookup * fix: eliminate configNameToKey race via caller-provided configServers param Replace the process-global configNameToKey map (last-writer-wins under concurrent multi-tenant load) with a configServers parameter on getServerConfig. Callers pass the pre-resolved config servers map directly — no shared mutable state, no cross-tenant race. - Add optional configServers param to getServerConfig; when provided, returns matching config directly without any global lookup - Remove configNameToKey map entirely (was the source of the race) - Extract server names from cache keys via lastIndexOf in invalidateConfigCache (safe for names containing colons) - Use mcpConfig[serverName] directly in getMCPTools instead of a redundant getServerConfig call - Add cross-tenant isolation test for getServerConfig * fix: populate read-through cache after config server lazy init After lazyInitConfigServer succeeds, write the parsed config to readThroughCache keyed by serverName so that getServerConfig calls from ConnectionsRepository, UserConnectionManager, and MCPManager.callTool find the config without needing configServers. Without this, config-source servers appeared in tool listings but every connection attempt and tool call returned undefined. * fix: user-scoped getServerConfig fallback to server-only cache key When getServerConfig is called with a userId (e.g., from callTool or UserConnectionManager), the cache key is serverName::userId. Config-source servers are cached under the server-only key (no userId). Add a fallback so user-scoped lookups find config-source servers in the read-through cache. * fix: configCacheRepo fallback, isUserSourced DRY, cross-process race CRITICAL: Add findInConfigCache fallback in getServerConfig so config-source servers remain reachable after readThroughCache TTL expires (5s). Without this, every tool call after 5s returned undefined for config-source servers. MAJOR: Extract isUserSourced() helper to mcp/utils.ts and replace all 5 inline dbSourced ternary expressions (MCPManager x2, ConnectionsRepository, UserConnectionManager, MCPServerInspector). MAJOR: Fix cross-process Redis race in lazyInitConfigServer — when configCacheRepo.add throws (key exists from another process), fall back to reading the existing entry instead of returning undefined. MINOR: Parallelize invalidateConfigCache awaits with Promise.all. Remove redundant .catch(() => {}) inside Promise.allSettled. Tighten dedup test assertion to toBe(1). Add TTL-expiry tests for getServerConfig (with and without userId). * feat: thread configServers through getAppToolFunctions and formatInstructionsForContext Add optional configServers parameter to getAppToolFunctions, getInstructions, and formatInstructionsForContext so config-source server tools and instructions are visible to agent initialization and context injection paths. Existing callers (boot-time init, tests) pass no argument and continue to work unchanged. Agent runtime paths can now thread resolved config servers from request context. * fix: stale failure stubs retry after 5 min, upsert for cross-process races - Add CONFIG_STUB_RETRY_MS (5 min) — stale failure stubs are retried instead of permanently disabling config-source servers after transient errors (DNS outage, cold-start race) - Extract upsertConfigCache() helper that tries add then falls back to update, preventing cross-process Redis races where a second instance's successful inspection result was discarded - Add test for stale-stub retry after CONFIG_STUB_RETRY_MS * fix: stamp updatedAt on failure stubs, null-guard callTool config, test cleanup - Add updatedAt: Date.now() to failure stubs in lazyInitConfigServer so CONFIG_STUB_RETRY_MS (5 min) window works correctly — without it, stubs were always considered stale (updatedAt ?? 0 → epoch → always expired) - Add null guard for rawConfig in MCPManager.callTool before passing to preProcessGraphTokens — prevents unsafe `as` cast on undefined - Log double-failure in upsertConfigCache instead of silently swallowing - Replace module-scope Date.now monkey-patch with jest.useFakeTimers / jest.setSystemTime / jest.useRealTimers in ensureConfigServers tests * fix: server-only readThrough fallback only returns truthy values Prevents a cached undefined from a prior no-userId lookup from short-circuiting the DB query on a subsequent userId-scoped lookup. * fix: remove findInConfigCache to eliminate cross-tenant config leakage The findInConfigCache prefix scan (serverName:*) could return any tenant's config after readThrough TTL expires, violating tenant isolation. Config-source servers are now ONLY resolvable through: 1. The configServers param (callers with tenant context from ALS) 2. The readThrough cache (populated by ensureSingleConfigServer, 5s TTL, repopulated on every HTTP request via resolveAllMcpConfigs) Connection/tool-call paths without tenant context rely exclusively on the readThrough cache. If it expires before the next HTTP request repopulates it, the server is not found — which is correct because there is no tenant context to determine which config to return. - Remove findInConfigCache method and its call in getServerConfig - Update server-only readThrough fallback to only return truthy values (prevents cached undefined from short-circuiting user-scoped DB lookup) - Update tests to document tenant isolation behavior after cache expiry * style: fix import order per AGENTS.md conventions Sort package imports shortest-to-longest, local imports longest-to-shortest across MCPServersRegistry, ConnectionsRepository, MCPManager, UserConnectionManager, and MCPServerInspector. * fix: eliminate cross-tenant readThrough contamination and TTL-expiry tool failures Thread pre-resolved serverConfig from tool creation context into callTool, removing dependency on the readThrough cache for config-source servers. This fixes two issues: - Cross-tenant contamination: the readThrough cache key was unscoped (just serverName), so concurrent multi-tenant requests for same-named servers would overwrite each other's entries - TTL expiry: tool calls happening >5s after config resolution would fail with "Configuration not found" because the readThrough entry had expired Changes: - Add optional serverConfig param to MCPManager.callTool — uses provided config directly, falling back to getServerConfig lookup for YAML/user servers - Thread serverConfig from createMCPTool through createToolInstance closure to callTool - Remove readThrough write from ensureSingleConfigServer — config-source servers are only accessible via configServers param (tenant-scoped) - Remove server-only readThrough fallback from getServerConfig - Increase config cache hash from 8 to 16 hex chars (64-bit) - Add isUserSourced boundary tests for all source/dbId combinations - Fix double Object.keys call in getMCPTools controller - Update test assertions for new getServerConfig behavior * fix: cache base configs for config-server users; narrow upsertConfigCache error handling - Refactor getAllServerConfigs to separate base config fetch (YAML + DB) from config-server layering. Base configs are cached via readThroughCacheAll regardless of whether configServers is provided, eliminating uncached MongoDB queries per request for config-server users - Narrow upsertConfigCache catch to duplicate-key errors only; infrastructure errors (Redis timeouts, network failures) now propagate instead of being silently swallowed, preventing inspection storms during outages * fix: restore correct merge order and document upsert error matching - Restore YAML → Config → User DB precedence in getAllServerConfigs (user DB servers have highest precedence, matching the JSDoc contract) - Add source comment on upsertConfigCache duplicate-key detection linking to the two cache implementations that define the error message * feat: complete config-source server support across all execution paths Wire configServers through the entire agent execution pipeline so config-source MCP servers are fully functional — not just visible in listings but executable in agent sessions. - Thread configServers into handleTools.js agent tool pipeline: resolve config servers from tenant context before MCP tool iteration, pass to getServerConfig, createMCPTools, and createMCPTool - Thread configServers into agent instructions pipeline: applyContextToAgent → getMCPInstructionsForServers → formatInstructionsForContext, resolved in client.js before agent context application - Add configServers param to createMCPTool and createMCPTools for reconnect path fallback - Add source field to redactServerSecrets allowlist for client UI differentiation of server tiers - Narrow invalidateConfigCache to only clear readThroughCacheAll (merged results), preserving YAML individual-server readThrough entries - Update context.spec.ts assertions for new configServers parameter * fix: add missing mocks for config-source server dependencies in client.test.js Mock getMCPServersRegistry, getAppConfig, and getTenantId that were added to client.js but not reflected in the test file's jest.mock declarations. * fix: update formatInstructionsForContext assertions for configServers param The test assertions expected formatInstructionsForContext to be called with only the server names array, but it now receives configServers as a second argument after the config-source server feature wiring. * fix: move configServers resolution before MCP tool loop to avoid TDZ configServers was declared with `let` after the first tool loop but referenced inside it via getServerConfig(), causing a ReferenceError temporal dead zone. Move declaration and resolution before the loop, using tools.some(mcpToolPattern) to gate the async resolution. * fix: address review findings — cache bypass, discoverServerTools gap, DRY - #2: getAllServerConfigs now always uses getBaseServerConfigs (cached via readThroughCacheAll) instead of bypassing it when configServers is present. Extracts user-DB entries from cached base by diffing against YAML keys to maintain YAML → Config → User DB merge order without extra MongoDB calls. - #3: Add configServers param to ToolDiscoveryOptions and thread it through discoverServerTools → getServerConfig so config-source servers are discoverable during OAuth reconnection flows. - #6: Replace inline import() type annotations in context.ts with proper import type { ParsedServerConfig } per AGENTS.md conventions. - #7: Extract resolveConfigServers(req) helper in MCP.js and use it from handleTools.js and client.js, eliminating the duplicated 6-line config resolution pattern. - #10: Restore removed "why" comment explaining getLoaded() vs getAll() choice in getMCPSetupData — documents non-obvious correctness constraint. - #11: Fix incomplete JSDoc param type on resolveAllMcpConfigs. * fix: consolidate imports, reorder constants, fix YAML-DB merge edge case - Merge duplicate @librechat/data-schemas requires in MCP.js into one - Move resolveConfigServers after module-level constants - Fix getAllServerConfigs edge case where user-DB entry overriding a YAML entry with the same name was excluded from userDbConfigs; now uses reference equality check to detect DB-overwritten YAML keys * fix: replace fragile string-match error detection with proper upsert method Add upsert() to IServerConfigsRepositoryInterface and all implementations (InMemory, Redis, RedisAggregateKey, DB). This eliminates the brittle error message string match ('already exists in cache') in upsertConfigCache that was the only thing preventing cross-process init races from silently discarding inspection results. Each implementation handles add-or-update atomically: - InMemory: direct Map.set() - Redis: direct cache.set() - RedisAggregateKey: read-modify-write under write lock - DB: delegates to update() (DB servers use explicit add() with ACL setup) * fix: wire configServers through remaining HTTP endpoints - getMCPServerById: use resolveAllMcpConfigs instead of bare getServerConfig - reinitialize route: resolve configServers before getServerConfig - auth-values route: resolve configServers before getServerConfig - getOAuthHeaders: accept configServers param, thread from callers - Update mcp.spec.js tests to mock getAllServerConfigs for GET by name * fix: thread serverConfig through getConnection for config-source servers Config-source servers exist only in configCacheRepo, not in YAML cache or DB. When callTool → getConnection → getUserConnection → getServerConfig runs without configServers, it returns undefined and throws. Fix by threading the pre-resolved serverConfig (providedConfig) from callTool through getConnection → getUserConnection → createUserConnectionInternal, using it as a fallback before the registry lookup. * fix: thread configServers through reinit, reconnect, and tool definition paths Wire configServers through every remaining call chain that creates or reconnects MCP server connections: - reinitMCPServer: accepts serverConfig and configServers, uses them for getServerConfig fallback, getConnection, and discoverServerTools - reconnectServer: accepts and passes configServers to reinitMCPServer - createMCPTools/createMCPTool: pass configServers to reconnectServer - ToolService.loadToolDefinitionsWrapper: resolves configServers from req, passes to both reinitMCPServer call sites - reinitialize route: passes serverConfig and configServers to reinitMCPServer * fix: address review findings — simplify merge, harden error paths, fix log labels - Simplify getAllServerConfigs merge: replace fragile reference-equality loop with direct spread { ...yamlConfigs, ...configServers, ...base } - Guard upsertConfigCache in lazyInitConfigServer catch block so cache failures don't mask the original inspection error - Deduplicate getYamlServerNames cold-start with promise dedup pattern - Remove dead `if (!mcpConfig)` guard in getMCPSetupData - Fix hardcoded "App server" in ServerConfigsCacheRedisAggregateKey error messages — now uses this.namespace for correct Config/App labeling - Remove misleading OAuth callback comment about readThrough cache - Move resolveConfigServers after module-level constants in MCP.js * fix: clear rejected yamlServerNames promise, fix config-source reinspect, fix reset log label - Clear yamlServerNamesPromise on rejection so transient cache errors don't permanently prevent ensureConfigServers from working - Skip reinspectServer for config-source servers (source: 'config') in reinitMCPServer — they lack a CACHE/DB storage location; retry is handled by CONFIG_STUB_RETRY_MS in ensureConfigServers - Use source field instead of dbId for storageLocation derivation - Fix remaining hardcoded "App" in reset() leaderCheck message * fix: persist oauthHeaders in flow state for config-source OAuth servers The OAuth callback route has no JWT auth context and cannot resolve config-source server configs. Previously, getOAuthHeaders would silently return {} for config-source servers, dropping custom token exchange headers. Now oauthHeaders are persisted in MCPOAuthFlowMetadata during flow initiation (which has auth context), and the callback reads them from the stored flow state with a fallback to the registry lookup for YAML/user-DB servers. * fix: update tests for getMCPSetupData null guard removal and ToolService mock - MCP.spec.js: update test to expect graceful handling of null mcpConfig instead of a throw (getAllServerConfigs always returns an object) - MCP.js: add defensive || {} for Object.entries(mcpConfig) in case of null from test mocks - ToolService.spec.js: add missing mock for ~/server/services/MCP (resolveConfigServers) * fix: address review findings — DRY, naming, logging, dead code, defensive guards - #1: Simplify getAllServerConfigs to single getBaseServerConfigs call, eliminating redundant double-fetch of cacheConfigsRepo.getAll() - #2: Add warning log when oauthHeaders absent from OAuth callback flow state - #3: Extract resolveAllMcpConfigs to MCP.js service layer; controller imports shared helper instead of reimplementing - #4: Rename _serverConfig/_provider to capturedServerConfig/capturedProvider in createToolInstance — these are actively used, not unused - #5: Log rejected results from ensureConfigServers Promise.allSettled so cache errors are visible instead of silently dropped - #6: Remove dead 'MCP config not found' error handlers from routes - #7: Document circular-dependency reason for dynamic require in clearMcpConfigCache - #8: Remove logger.error from withTimeout to prevent double-logging timeouts - #10: Add explicit userId guard in ServerConfigsDB.upsert with clear error message - #12: Use spread instead of mutation in addServer for immutability consistency - Add upsert mock to ensureConfigServers.test.ts DB mock - Update route tests for resolveAllMcpConfigs import change * fix: restore correct merge priority, use immutable spread, fix test mock - getAllServerConfigs: { ...configServers, ...base } so userDB wins over configServers, matching documented "User DB (highest)" priority - lazyInitConfigServer: use immutable spread instead of direct mutation for parsedConfig.source, consistent with addServer fix - Fix test to mock getAllServerConfigs as {} instead of null, remove unnecessary || {} defensive guard in getMCPSetupData * fix: error handling, stable hashing, flatten nesting, remove dead param - Wrap resolveConfigServers/resolveAllMcpConfigs in try/catch with graceful {} fallback so transient DB/cache errors don't crash tool pipeline - Sort keys in configCacheKey JSON.stringify for deterministic hashing regardless of object property insertion order - Flatten clearMcpConfigCache from 3 nested try-catch to early returns; document that user connections are cleaned up lazily (accepted tradeoff) - Remove dead configServers param from getAppToolFunctions (never passed) - Add security rationale comment for source field in redactServerSecrets * fix: use recursive key-sorting replacer in configCacheKey to prevent cross-tenant cache collision The array replacer in JSON.stringify acts as a property allowlist at every nesting depth, silently dropping nested keys like headers['X-API-Key'], oauth.client_secret, etc. Two configs with different nested values but identical top-level structure produced the same hash, causing cross-tenant cache hits and potential credential contamination. Switch to a function replacer that recursively sorts keys at all depths without dropping any properties. Also document the known gap in getOAuthServers: config-source OAuth servers are not covered by auto-reconnection or uninstall cleanup because callers lack request context. * fix: move clearMcpConfigCache to packages/api to eliminate circular dependency The function only depends on MCPServersRegistry and MCPManager, both of which live in packages/api. Import it directly from @librechat/api in the CJS layer instead of using dynamic require('~/config'). * chore: imports/fields ordering * fix: address review findings — error handling, targeted lookup, test gaps - Narrow resolveAllMcpConfigs catch to only wrap ensureConfigServers so getAppConfig/getAllServerConfigs failures propagate instead of masking infrastructure errors as empty server lists. - Use targeted getServerConfig in getMCPServerById instead of fetching all server configs for a single-server lookup. - Forward configServers to inner createMCPTool calls so reconnect path works for config-source servers. - Update getAllServerConfigs JSDoc to document disjoint-key design. - Add OAuth callback oauthHeaders fallback tests (flow state present vs registry fallback). - Add resolveConfigServers/resolveAllMcpConfigs unit tests covering happy path and error propagation. * fix: add getOAuthReconnectionManager mock to OAuth callback tests * chore: imports ordering	2026-03-28 10:36:43 -04:00
Danny Avila	6f87b49df8	🛂 fix: Enforce Actions Capability Gate Across All Event-Driven Tool Loading Paths (#12252) ... * fix: gate action tools by actions capability in all code paths Extract resolveAgentCapabilities helper to eliminate 3x-duplicated capability resolution. Apply early action-tool filtering in both loadToolDefinitionsWrapper and loadAgentTools non-definitions path. Gate loadActionToolsForExecution in loadToolsForExecution behind an actionsEnabled parameter with a cache-based fallback. Replace the late capability guard in loadAgentTools with a hasActionTools check to avoid unnecessary loadActionSets DB calls and duplicate warnings. * fix: thread actionsEnabled through InitializedAgent type Add actionsEnabled to the loadTools callback return type, InitializedAgent, and the initializeAgent destructuring/return so callers can forward the resolved value to loadToolsForExecution without redundant getEndpointsConfig cache lookups. * fix: pass actionsEnabled from callers to loadToolsForExecution Thread actionsEnabled through the agentToolContexts map in initialize.js (primary and handoff agents) and through primaryConfig in the openai.js and responses.js controllers, avoiding per-tool-call capability re-resolution on the hot path. * test: add regression tests for action capability gating Test the real exported functions (resolveAgentCapabilities, loadAgentTools, loadToolsForExecution) with mocked dependencies instead of shadow re-implementations. Covers definition filtering, execution gating, actionsEnabled param forwarding, and fallback capability resolution. * test: use Constants.EPHEMERAL_AGENT_ID in ephemeral fallback test Replaces a string guess with the canonical constant to avoid fragility if the ephemeral detection heuristic changes. * fix: populate agentToolContexts for addedConvo parallel agents After processAddedConvo returns, backfill agentToolContexts for any agents in agentConfigs not already present, so ON_TOOL_EXECUTE for added-convo agents receives actionsEnabled instead of falling back to a per-call cache lookup.	2026-03-15 23:01:36 -04:00
Danny Avila	d3c06052d7	🗝️ feat: Credential Variables for DB-Sourced MCP Servers (#12044) ... * feat: Allow Credential Variables in Headers for DB-sourced MCP Servers - Removed the hasCustomUserVars check from ToolService.js, directly retrieving userMCPAuthMap. - Updated MCPConnectionFactory and related classes to include a dbSourced flag for better handling of database-sourced configurations. - Added integration tests to ensure proper behavior of dbSourced servers, verifying that sensitive placeholders are not resolved while allowing customUserVars. - Adjusted various MCP-related files to accommodate the new dbSourced logic, ensuring consistent handling across the codebase. * chore: MCPConnectionFactory Tests with Additional Flow Metadata for typing - Updated MCPConnectionFactory tests to include new fields in flowMetadata: serverUrl and state. - Enhanced mockFlowData in multiple test cases to reflect the updated structure, ensuring comprehensive coverage of the OAuth flow scenarios. - Added authorization_endpoint to metadata in the test setup for improved validation of the OAuth process. * refactor: Simplify MCPManager Configuration Handling - Removed unnecessary type assertions and streamlined the retrieval of server configuration in MCPManager. - Enhanced the handling of OAuth and database-sourced flags for improved clarity and efficiency. - Updated tests to reflect changes in user object structure and ensure proper processing of MCP environment variables. * refactor: Optimize User MCP Auth Map Retrieval in ToolService - Introduced conditional loading of userMCPAuthMap based on the presence of MCP-delimited tools, improving efficiency by avoiding unnecessary calls. - Updated the loadToolDefinitionsWrapper and loadAgentTools functions to reflect this change, enhancing overall performance and clarity. * test: Add userMCPAuthMap gating tests in ToolService - Introduced new tests to validate the logic for determining if MCP tools are present in the agent's tool list. - Implemented various scenarios to ensure accurate detection of MCP tools, including edge cases for empty, undefined, and null tool lists. - Enhanced clarity and coverage of the ToolService capability checking logic. * refactor: Enhance MCP Environment Variable Processing - Simplified the handling of the dbSourced parameter in the processMCPEnv function. - Introduced a failsafe mechanism to derive dbSourced from options if not explicitly provided, improving robustness and clarity in MCP environment variable processing. * refactor: Update Regex Patterns for Credential Placeholders in ServerConfigsDB - Modified regex patterns to include additional credential/env placeholders that should not be allowed in user-provided configurations. - Clarified comments to emphasize the security risks associated with credential exfiltration when MCP servers are shared between users. * chore: field order * refactor: Clean Up dbSourced Parameter Handling in processMCPEnv - Reintroduced the failsafe mechanism for deriving the dbSourced parameter from options, ensuring clarity and robustness in MCP environment variable processing. - Enhanced code readability by maintaining consistent comment structure. * refactor: Update MCPOptions Type to Include Optional dbId - Modified the processMCPEnv function to extend the MCPOptions type, allowing for an optional dbId property. - Simplified the logic for deriving the dbSourced parameter by directly checking the dbId property, enhancing code clarity and maintainability.	2026-03-03 18:02:37 -05:00
Danny Avila	3d98194a99	🦥 feat: Add Deferred Tools as Agents Capability (#11295)	2026-01-28 17:44:30 -05:00