abhigyanpatwari/GitNexus · critical

-32000

-32000

Error message

Internal MCP server error

What it means

JSON-RPC error response (code -32000, message 'Internal MCP server error', id null) returned by the /api/mcp streamable-HTTP endpoint when the MCP handler itself throws or rejects. It is the MCP surface's equivalent of an unhandled exception: the real failure is logged as 'MCP HTTP request failed:'; if headers were already sent (failure mid-stream) no error body can be written and the HTTP stream simply terminates.

Source

Thrown at gitnexus/src/server/mcp-http.ts:28

import type { Express, Request, Response } from 'express';
import { createStreamableHttpHandler } from '../mcp/http-transport.js';
import type { LocalBackend } from '../mcp/local/local-backend.js';
import { createMcpRepositoryPolicy } from '../mcp/repository-policy.js';
import { logger } from '../core/logger.js';

export async function mountMCPEndpoints(
  app: Express,
  backend: LocalBackend,
): Promise<() => Promise<void>> {
  const repositoryPolicy = await createMcpRepositoryPolicy(backend);
  const { handler, cleanup } = createStreamableHttpHandler(backend, { repositoryPolicy });

  app.all('/api/mcp', (req: Request, res: Response) => {
    void handler(req, res).catch((err: unknown) => {
      logger.error({ err }, 'MCP HTTP request failed:');
      if (res.headersSent) return;
      res.status(500).json({
        jsonrpc: '2.0',
        error: { code: -32000, message: 'Internal MCP server error' },
        id: null,
      });
    });
  });

  logger.info('MCP HTTP endpoints mounted at /api/mcp');
  return cleanup;
}

View on GitHub (pinned to aac7515d2a)

Solutions

1. Check the serve logs for 'MCP HTTP request failed:' — the -32000 body carries no detail by design
2. Let the client re-initialize the MCP session (most do automatically) and retry the tool call once
3. Re-run analyze to rebuild the index if storage corruption is suspected
4. Persistent -32000 with a reproducible request is a bug — update GitNexus and report with the stack

Example fix

// before
const result = await client.callTool({ name: 'impact', arguments: args }); // throws on -32000

// after
async function callMcpSafe(client: any, name: string, args: object) {
  try {
    return await client.callTool({ name, arguments: args });
  } catch (e: any) {
    if (e?.code === -32000) {
      await client.connect(); // re-initialize the session
      return client.callTool({ name, arguments: args });
    }
    throw e;
  }
}

Catch JSON-RPC errors with code -32000, log context client-side, re-initialize the MCP session, and retry the call exactly once; escalate to the serve log if it persists.

When it happens

Trigger: A backend crash while servicing tools/list or tools/call; corrupted or version-skewed index/storage touched during a tool call; a GitNexus bug in the streamable-HTTP handler; the server dying mid-request and leaving a half-written response.

Common situations: Editor/MCP clients attached to a serve instance with corrupted storage; concurrent MCP sessions racing index re-initialization; upgrades leaving the index in an incompatible state.

Related errors

• Failed to start analysis
• Failed to start embedding generation
• Missing Content-Length header from MCP client
• Invalid Content-Length header from MCP client
• Content-Length ${contentLength} exceeds maximum allowed size

AI-assisted analysis of abhigyanpatwari/GitNexus@aac7515d2a (2026-08-20). Data as JSON: /api/errors/3a3f8f56745e9a5b. Report an issue: GitHub.