Hi,
I’m trying to create an OpenAI Secure MCP Tunnel for a local MCP server that I want to use from a ChatGPT custom connector.
My intended setup is:
ChatGPT custom connector
-> Connection = Tunnel
-> OpenAI Secure MCP Tunnel
-> local tunnel-client
-> local MCP server at http://127.0.0.1:8000/mcp
I understand that ChatGPT does not accept a direct local HTTP MCP URL like http://localhost:8000/mcp, so I’m trying to use the Tunnel option instead.
In Platform settings, I created two custom roles:
-
Tunnel Manager:
Tunnels Read + Manage -
Tunnel User:
Tunnels Read + Use
I assigned both roles to my user.
Then I went to Platform → Organization → Tunnels and tried to create a tunnel with:
Name: llmgw
Description: local MCP tunnel
Organization IDs: my current org ID
Workspace IDs: empty
The UI shows a generic toast:
Failed to create tunnel. Please try again.
The Network response contains:
{
"error": {
"message": "403: Tenant context missing and no allowlist is configured for the active organization. (reason=fallback_missing_entry) (request_id=req_09e00d1c231347b48dcc8a5ed305605f)",
"type": "server_error",
"param": null,
"code": null
}
}
This looks somewhat related to another private MCP connector issue mentioning a missing access_list, but in my case the backend says fallback_missing_entry.
Questions:
-
Is Secure MCP Tunnel currently available for personal Platform organizations / personal ChatGPT accounts?
-
Is a ChatGPT Workspace ID required to create a tunnel?
-
If yes, where can I find the correct
workspace-...ID for a personal ChatGPT workspace? -
Does
fallback_missing_entrymean my organization needs to be allowlisted before Secure MCP Tunnel can be used? -
What is the expected relationship between Platform organizations, Platform projects, ChatGPT workspaces, and Secure MCP Tunnel scope?
I’m not trying to report a billing issue. I’m trying to understand the correct Apps SDK / MCP tunnel setup for local development with ChatGPT connectors.
Thanks!
