LibreChat/api/server/services/Tools/mcp.js
Danny Avila 32cadb1cc5
Some checks are pending
Docker Dev Branch Images Build / build (Dockerfile, lc-dev, node) (push) Waiting to run
Docker Dev Branch Images Build / build (Dockerfile.multi, lc-dev-api, api-build) (push) Waiting to run
Docker Dev Images Build / build (Dockerfile, librechat-dev, node) (push) Waiting to run
Docker Dev Images Build / build (Dockerfile.multi, librechat-dev-api, api-build) (push) Waiting to run
Sync Locize Translations & Create Translation PR / Sync Translation Keys with Locize (push) Waiting to run
Sync Locize Translations & Create Translation PR / Create Translation PR on Version Published (push) Blocked by required conditions
🩹 fix: MCP Server Recovery from Startup Inspection Failures (#12145)
* feat: MCP server reinitialization recovery mechanism

- Added functionality to store a stub configuration for MCP servers that fail inspection at startup, allowing for recovery via reinitialization.
- Introduced `reinspectServer` method in `MCPServersRegistry` to handle reinspection of previously failed servers.
- Enhanced `MCPServersInitializer` to log and manage server initialization failures, ensuring proper handling of inspection failures.
- Added integration tests to verify the recovery process for unreachable MCP servers, ensuring that stub configurations are stored and can be reinitialized successfully.
- Updated type definitions to include `inspectionFailed` flag in server configurations for better state management.

* fix: MCP server handling for inspection failures

- Updated `reinitMCPServer` to return a structured response when the server is unreachable, providing clearer feedback on the failure.
- Modified `ConnectionsRepository` to prevent connections to servers marked as inspection failed, improving error handling.
- Adjusted `MCPServersRegistry` methods to ensure proper management of server states, including throwing errors for non-failed servers during reinspection.
- Enhanced integration tests to validate the behavior of the system when dealing with unreachable MCP servers and inspection failures, ensuring robust recovery mechanisms.

* fix: Clear all cached server configurations in MCPServersRegistry

- Added a comment to clarify the necessity of clearing all cached server configurations when updating a server's configuration, as the cache is keyed by userId without a reverse index for enumeration.

* fix: Update integration test for file_tools_server inspection handling

- Modified the test to verify that the `file_tools_server` is stored as a stub when inspection fails, ensuring it can be reinitialized correctly.
- Adjusted expectations to confirm that the `inspectionFailed` flag is set to true for the stub configuration, enhancing the robustness of the recovery mechanism.

* test: Add unit tests for reinspecting servers in MCPServersRegistry

- Introduced tests for the `reinspectServer` method to validate error handling when called on a healthy server and when the server does not exist.
- Ensured that appropriate exceptions are thrown for both scenarios, enhancing the robustness of server state management.

* test: Add integration test for concurrent reinspectServer calls

- Introduced a new test to validate that multiple concurrent calls to reinspectServer do not crash or corrupt the server state.
- Ensured that at least one call succeeds and any failures are due to the server not being in a failed state, enhancing the reliability of the reinitialization process.

* test: Enhance integration test for concurrent MCP server reinitialization

- Added a new test to validate that concurrent calls to reinitialize the MCP server do not crash or corrupt the server state.
- Ensured that at least one call succeeds and that failures are handled gracefully, improving the reliability of the reinitialization process.
- Reset MCPManager instance after each test to maintain a clean state for subsequent tests.
2026-03-08 21:49:04 -04:00

210 lines
6.8 KiB
JavaScript

const { logger } = require('@librechat/data-schemas');
const { CacheKeys, Constants } = require('librechat-data-provider');
const { getMCPManager, getMCPServersRegistry, getFlowStateManager } = require('~/config');
const { findToken, createToken, updateToken, deleteTokens } = require('~/models');
const { updateMCPServerTools } = require('~/server/services/Config');
const { getLogStores } = require('~/cache');
/**
* Reinitializes an MCP server connection and discovers available tools.
* When OAuth is required, uses discovery mode to list tools without full authentication
* (per MCP spec, tool listing should be possible without auth).
* @param {Object} params
* @param {IUser} params.user - The user from the request object.
* @param {string} params.serverName - The name of the MCP server
* @param {boolean} params.returnOnOAuth - Whether to initiate OAuth and return, or wait for OAuth flow to finish
* @param {AbortSignal} [params.signal] - The abort signal to handle cancellation.
* @param {boolean} [params.forceNew]
* @param {number} [params.connectionTimeout]
* @param {FlowStateManager<any>} [params.flowManager]
* @param {(authURL: string) => Promise<void>} [params.oauthStart]
* @param {Record<string, Record<string, string>>} [params.userMCPAuthMap]
*/
async function reinitMCPServer({
user,
signal,
forceNew,
serverName,
userMCPAuthMap,
connectionTimeout,
returnOnOAuth = true,
oauthStart: _oauthStart,
flowManager: _flowManager,
}) {
/** @type {MCPConnection | null} */
let connection = null;
/** @type {LCAvailableTools | null} */
let availableTools = null;
/** @type {ReturnType<MCPConnection['fetchTools']> | null} */
let tools = null;
let oauthRequired = false;
let oauthUrl = null;
try {
const registry = getMCPServersRegistry();
const serverConfig = await registry.getServerConfig(serverName, user?.id);
if (serverConfig?.inspectionFailed) {
logger.info(
`[MCP Reinitialize] Server ${serverName} had failed inspection, attempting reinspection`,
);
try {
const storageLocation = serverConfig.dbId ? 'DB' : 'CACHE';
await registry.reinspectServer(serverName, storageLocation, user?.id);
logger.info(`[MCP Reinitialize] Reinspection succeeded for server: ${serverName}`);
} catch (reinspectError) {
logger.error(
`[MCP Reinitialize] Reinspection failed for server ${serverName}:`,
reinspectError,
);
return {
availableTools: null,
success: false,
message: `MCP server '${serverName}' is still unreachable`,
oauthRequired: false,
serverName,
oauthUrl: null,
tools: null,
};
}
}
const customUserVars = userMCPAuthMap?.[`${Constants.mcp_prefix}${serverName}`];
const flowManager = _flowManager ?? getFlowStateManager(getLogStores(CacheKeys.FLOWS));
const mcpManager = getMCPManager();
const tokenMethods = { findToken, updateToken, createToken, deleteTokens };
const oauthStart =
_oauthStart ??
(async (authURL) => {
logger.info(`[MCP Reinitialize] OAuth URL received for ${serverName}`);
oauthUrl = authURL;
oauthRequired = true;
});
try {
connection = await mcpManager.getConnection({
user,
signal,
forceNew,
oauthStart,
serverName,
flowManager,
tokenMethods,
returnOnOAuth,
customUserVars,
connectionTimeout,
});
logger.info(`[MCP Reinitialize] Successfully established connection for ${serverName}`);
} catch (err) {
logger.info(`[MCP Reinitialize] getConnection threw error: ${err.message}`);
logger.info(
`[MCP Reinitialize] OAuth state - oauthRequired: ${oauthRequired}, oauthUrl: ${oauthUrl ? 'present' : 'null'}`,
);
const isOAuthError =
err.message?.includes('OAuth') ||
err.message?.includes('authentication') ||
err.message?.includes('401');
const isOAuthFlowInitiated = err.message === 'OAuth flow initiated - return early';
if (isOAuthError || oauthRequired || isOAuthFlowInitiated) {
logger.info(
`[MCP Reinitialize] OAuth required for ${serverName}, attempting tool discovery without auth`,
);
oauthRequired = true;
try {
const discoveryResult = await mcpManager.discoverServerTools({
user,
signal,
serverName,
flowManager,
tokenMethods,
oauthStart,
customUserVars,
connectionTimeout,
});
if (discoveryResult.tools && discoveryResult.tools.length > 0) {
tools = discoveryResult.tools;
logger.info(
`[MCP Reinitialize] Discovered ${tools.length} tools for ${serverName} without full auth`,
);
}
} catch (discoveryErr) {
logger.debug(
`[MCP Reinitialize] Tool discovery failed for ${serverName}: ${discoveryErr?.message ?? String(discoveryErr)}`,
);
}
} else {
logger.error(
`[MCP Reinitialize] Error initializing MCP server ${serverName} for user:`,
err,
);
}
}
if (connection && !oauthRequired) {
tools = await connection.fetchTools();
}
if (tools && tools.length > 0) {
availableTools = await updateMCPServerTools({
userId: user.id,
serverName,
tools,
});
}
logger.debug(
`[MCP Reinitialize] Sending response for ${serverName} - oauthRequired: ${oauthRequired}, oauthUrl: ${oauthUrl ? 'present' : 'null'}`,
);
const getResponseMessage = () => {
if (oauthRequired && tools && tools.length > 0) {
return `MCP server '${serverName}' tools discovered, OAuth required for execution`;
}
if (oauthRequired) {
return `MCP server '${serverName}' ready for OAuth authentication`;
}
if (connection) {
return `MCP server '${serverName}' reinitialized successfully`;
}
return `Failed to reinitialize MCP server '${serverName}'`;
};
const result = {
availableTools,
success: Boolean(
(connection && !oauthRequired) ||
(oauthRequired && oauthUrl) ||
(tools && tools.length > 0),
),
message: getResponseMessage(),
oauthRequired,
serverName,
oauthUrl,
tools,
};
logger.debug(`[MCP Reinitialize] Response for ${serverName}:`, {
success: result.success,
oauthRequired: result.oauthRequired,
oauthUrl: result.oauthUrl ? 'present' : null,
toolsCount: tools?.length ?? 0,
});
return result;
} catch (error) {
logger.error(
'[MCP Reinitialize] Error loading MCP Tools, servers may still be initializing:',
error,
);
}
}
module.exports = {
reinitMCPServer,
};