{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreietn5nshqq5m22ioj3mnserx6xxyg6pmqubmtobyxmv7rt7zrqgh4",
    "uri": "at://did:plc:lk3jfj3zq4k4wxnk474axylu/app.bsky.feed.post/3moowzhups2t2"
  },
  "path": "/t/mcp-server-the-tool-registry-returned-an-error-saying-no-tools-were-defined/1383149#post_4",
  "publishedAt": "2026-06-20T03:32:45.000Z",
  "site": "https://community.openai.com",
  "textContent": "I would debug this as a discovery/registration problem first, not a tool implementation problem.\n\nA quick checklist:\n\n  1. Confirm the exact URL ChatGPT is using, including the path. If your server exposes tools under `/mcp`, do not test only the base host.\n  2. Compare the discovery response that your API client sees with what the ChatGPT app registration points at. It is possible for the server to be healthy while the app points at an old path.\n  3. Change one visible tool name or description and republish/rescan the app. If ChatGPT still shows the old metadata or no tools, you are looking at stale registration/cache rather than runtime execution.\n  4. Add request logging at the reverse proxy edge. If ChatGPT never hits the server during scan, the failure is before MCP runtime.\n  5. Make sure the published app version and the developer/test version are not using different connector URLs.\n\n\n\nThe useful signal is whether a scan produces any inbound request at all. No request means registration/path/auth. Request with an empty tools result means server discovery/schema.",
  "title": "MCP server: the tool registry returned an error saying no tools were defined"
}