Hi all — we’re running into a connector creation timeout that appears to be client-side / platform-side rather than server-side, and wanted to check if others are seeing the same.
Summary
When adding a custom MCP server as a connector in ChatGPT, the UI errors with “Error creating connector — Request timeout”, even though the MCP server successfully handles all expected calls.
What we’re seeing in server logs
-
Requests from User-Agent:
openai-mcp/1.0.0 -
initialize→ 200 OK in ~0–32 ms -
tools/list→ 200 OK, full tool list returned -
No redirects (301/302)
-
HTTPS valid
-
No errors, crashes, or restarts
-
CSP / CORS explicitly allow
chat.openai.comandchatgpt.com
What we’ve verified
-
MCP endpoint is mounted exactly at
/mcp -
No trailing slash redirects
-
Curl tests succeed:
-
POST /tools/list -
Requests with
Accept: text/event-stream
-
-
Railway service is warm (no cold-start delay)
-
Tool schemas and outputs are valid JSON
-
Same behavior across retries
Despite this, the ChatGPT UI still times out during connector creation.
Hypothesis
It seems the MCP handshake succeeds, but the UI may be timing out on a post-validation step (internal safety scan, tool eligibility check, caching, or connector metadata processing), rather than on the MCP server itself.
We’ve seen similar reports where:
-
initialize/tools/list succeed
-
UI still times out or fails creation
-
behavior appears intermittent or rollout-related
Questions
-
Is anyone else seeing connector creation timeouts even when MCP calls are fast and correct?
-
Are there known limits around:
-
tool count
-
schema size
-
connector metadata processing
-
-
Is this a known incident or regression with MCP connectors / Apps SDK right now?
Happy to provide sanitized logs or try any suggested mitigations (minimal tool mode, schema reduction, etc.).
Thanks in advance — appreciate any confirmation or guidance.