Github MCP server doesn't work inside ChatGPT

[ashot-khulaian]

I connected GitHub MCP (https://api.githubcopilot.com/mcp, not the default GH connector) to ChatGPT and am trying to create an issue in the repo. All permissions are set, and MCP can read the repo. But when creating an issue, ChatGPT fails with the error “Error in message stream”—no error details. What can be wrong?

[ashot-khulaian]

I asked OpenAI first, their reply was that it's MCP server error:

Thank you for sharing the screenshot. This matches the typical error message when there’s a problem with the message exchange between ChatGPT and your custom MCP connector. Since there are no extra error details shown in the app, the root cause is almost always on the MCP connector side.

Here’s what you should check next:

1. Review your MCP server’s logs during the failed request to see if it provides any error messages.

2. Make sure your MCP connector’s responses follow the official protocol—especially around the “create issue” tool, streaming, and message format.

3. Confirm that all HTTP endpoints used by your MCP are reachable from OpenAI and returning the expected payload/schema.

4. Double-check developer mode and tool permissions in ChatGPT settings.

If you’d like, you can share relevant MCP response examples (redacting sensitive info), and I can help check for schema or formatting issues. Let me know if you need further guidance on any step!

Hello, I’m an AI support agent for OpenAI. How can I help you with OpenAI products or services today?

You said:
I connected GitHub MCP (https: //api .githubcopilot.com/mcp, not the default GH connector) to ChatGPT and am trying to create an issue in the repo. All permissions are set, and MCP can read the repo. But when creating an issue, ChatGPT fails with the error “Error in message stream”—no error details. What can be wrong?
The assistant said:
Based on available information, the “Error in message stream” when using a custom GitHub MCP connector (not the default GitHub connector) usually means there is a compatibility issue between ChatGPT and the MCP server or the connector does not fully meet the required MCP protocol.
Common causes and steps to check:
Ensure the MCP server fully implements OpenAI's MCP specification. Even small missing pieces (like required tools or message structure) can cause generic errors. Check the full MCP documentation: https://platform.openai.com/docs/mcp
Confirm developer mode is enabled in ChatGPT settings if you're using a Business, Teams, or Pro seat.
Make sure the connector is published (not just in test/draft) and you have the right permissions.
Sometimes access permissions or allowlisting (ex: IP/domain) rules are missed, so review those too.
This error is not repo access related; it's typically a connector implementation or configuration issue openai.com.
If these checks don’t resolve it, the problem is most likely with the MCP connector’s implementation, and reviewing your server’s logs for errors during the “create issue” call could provide more detail.

The assistant said:
Thank you for sharing the screenshot. This matches the typical error message when there’s a problem with the message exchange between ChatGPT and your custom MCP connector. Since there are no extra error details shown in the app, the root cause is almost always on the MCP connector side.
Here’s what you should check next:
Review your MCP server’s logs during the failed request to see if it provides any error messages.
Make sure your MCP connector’s responses follow the official protocol—especially around the “create issue” tool, streaming, and message format.
Confirm that all HTTP endpoints used by your MCP are reachable from OpenAI and returning the expected payload/schema.
Double-check developer mode and tool permissions in ChatGPT settings.
If you’d like, you can share relevant MCP response examples (redacting sensitive info), and I can help check for schema or formatting issues. Let me know if you need further guidance on any step!

You said:
How can I review github mcp server logs
The assistant said:
To review your GitHub MCP server logs, you typically need access to the server or environment where your MCP connector is running. Here’s how you can do it for most setups:
Check the Server Hosting MCP: Access the virtual machine, container, or server where your MCP server code is running.
Review Log Files: Look for standard output/error logs. These are often written to files (like logs/, stdout.log, or service-specific log locations), or available via system tools (like journalctl for Linux services).
Enable Verbose Logging: If your MCP implementation supports increased logging (e.g., debug or verbose mode), enable it for more detailed error messages.
Cloud Providers: If hosting on cloud platforms (AWS, Azure, GCP), check their logging services (CloudWatch, Stackdriver, etc.).
Local Development: If running MCP locally (ex: using Node.js or Python), check your console or configure your app to save logs to a file.
Look for logs that are timed with your “create issue” attempt—these often show the request payload, any internal errors, or protocol mismatches.
If you need guidance on enabling or finding logs in your specific MCP server environment, let me know what tech stack you're using (e.g., Node.js, Python, Docker, etc.), and I can provide more targeted instructions.

You said:
I have no clue its a github mcp server
The assistant said:
The GitHub MCP connector you are using is hosted and operated by the organization or provider who maintains the MCP server. OpenAI does not operate or have direct access to these logs. To review logs for this MCP server, you would need to contact your internal IT team, system administrator, or whoever set up and manages the MCP endpoint (https://api.githubcopilot.com/mcp).
Generally, the steps are:
Locate the server, VM, or service that runs your MCP connector.
Check its log collection method (local files, cloud logging, container logs, etc.).
Review logs for errors at the time you try to create a GitHub issue via ChatGPT.
If you don’t know who manages this MCP server or have no admin access, you will need to reach out within your organization or IT department for support. Only whoever controls that server can retrieve those logs or add debug logging if needed.
For reference, the “Error in message stream” usually means the connector is not responding as required, or its responses do not match OpenAI’s MCP spec openai.com.

[mattdholloway]

Hi! I've tried this and have been unable to reproduce the error. If you attempt this again, does this error still occur?

[ashot-khulaian]

I aslo can't reproduce now, it works. Closing the issue.