ChatGPT Pro consistently routes DevSpace actions to wrong/disabled tools after connector setup

[pyro17ai]

Summary

I am using DevSpace as a ChatGPT MCP connector. The connector works with other ChatGPT models, but ChatGPT Pro has not successfully used DevSpace even once in my setup.

With Pro, the failure is consistent: instead of calling DevSpace tools such as open_workspace, read, bash, edit, etc., it repeatedly tries to call unrelated or disabled internal tools like multi_tool_use.parallel. Then it says that the required DevSpace tools are not available in its environment.

This looks like a ChatGPT Pro tool-router / active tool registry issue, but DevSpace may be able to mitigate it with stronger MCP instructions and clearer recovery guidance.

Environment

• Package: @waishnav/devspace
• Version: 1.0.1
• OS: Windows
• MCP host: ChatGPT connector
• Failing model/mode: ChatGPT Pro
• Working models: other ChatGPT models with the same DevSpace connector
• DevSpace exposed over HTTPS tunnel to /mcp

What happens

The same DevSpace connector works with other models, but in ChatGPT Pro:

1. DevSpace is connected as an MCP connector.
2. The model is asked to use DevSpace for local workspace access.
3. Pro either fails to call the right DevSpace tools or starts with DevSpace and then routes follow-up actions to the wrong tools.
4. It repeatedly attempts to call multi_tool_use.parallel, even though that tool does not exist or is disabled in the active environment.
5. It then claims that the required DevSpace tools are not available in its environment.

Example error:

"multi_tool_use tool is disabled. Do not send any messages to it."

In practice, Pro keeps trying to use multi_tool_use.parallel for DevSpace-related actions instead of using actual DevSpace MCP tools.

Expected behavior

After DevSpace is connected, ChatGPT Pro should use only the DevSpace MCP tools exposed by the connector.

Expected DevSpace workflow:

1. Call open_workspace once for the requested project folder.
2. Reuse the returned workspaceId.
3. Use DevSpace tools for all follow-up actions:
   • read / read_file
   • bash / run_shell
   • edit / edit_file
   • write / write_file
   • optional grep, glob, ls, show_changes if available
4. Never use unrelated internal tools such as multi_tool_use.parallel or container.exec for DevSpace workspace actions.

If the required DevSpace tools are not visible in the active tool registry, the model should stop and clearly report which DevSpace tools are available instead of attempting fallback calls to disabled tools.

Actual behavior

ChatGPT Pro consistently routes DevSpace-related actions to unavailable or disabled tools.

The most common wrong tool is:

multi_tool_use.parallel

The result is that Pro reports tool errors or says it does not have the required tools, even though the same connector works with other ChatGPT models.

This makes it look like DevSpace is broken, but the behavior strongly suggests that ChatGPT Pro is choosing the wrong tool layer.

Why I think this is a tool-router issue

The important detail is that the problem is specific to ChatGPT Pro.

The same DevSpace setup works with other models, but Pro repeatedly fails in the same way:

• calls the wrong tools;
• says required DevSpace tools are missing;
• tries to use multi_tool_use.parallel;
• never successfully completes a DevSpace workflow.

So this may not be a DevSpace server bug directly. It may be that Pro's tool router does not reliably follow the DevSpace MCP tool instructions.

Impact

This is a serious UX issue for ChatGPT Pro users.

The connector can be correctly configured and work in other models, but Pro still fails because it routes DevSpace actions into unavailable internal tools. Stronger DevSpace instructions and clearer failure handling could make this much less confusing and possibly prevent the wrong-tool calls.

[Waishnav]

GPT 5.5 pro doesn't support MCP tool calls for some accounts and for some account it is enabled.
I still don't know on what basis ChatGPT team has decided this, so far what i've figured is that if you have business account then it's enabled

i've also wrote public feature request proposal to em, you call follow there and check if any oai employees reach to it
https://x.com/wshxnv/status/2067641394832023748?s=20

[pyro17ai]

Thanks for your involvement in the project, bro! I use Pro all the time to make plans and do code reviews—I just zip up the working repo and drop it straight into ChatGPT Pro. So your project caught my eye, and I really hope the Codex team gets back to you soon.