openai / openai/codex

[Desktop] MCP tool calls fail with local models while the same MCP works with GPT-5.6

Open
#38,490 1 comment 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

app bug custom-model mcp tool-calls
Dominant language
Rust
Stars
125k
Forks
19.5k
PR merge metrics
PR metrics pending

Description

What version of the Codex App are you using (From “About Codex” dialog)?

26.810.41047

What subscription do you have?

free tier, i use ollama for work

What platform is your computer?

Darwin 25.4.0 arm64 arm

What issue are you seeing?

Description

I am experiencing an MCP tool-calling issue when using a local model in the Codex Desktop app.
The MCP server is discovered correctly and all of its tools are visible in the tool list, but calling an MCP tool with the local model fails with:
unsupported call
the exact same MCP configuration works correctly when I switch to GPT-5.6.

Environment

  • OS: macOS
  • Codex Desktop: 26.810.41047
  • Local model: Qwen3.6-35B-A3B、ornith 35B-A3B
  • MCP server: Brave Search MCP
  • Provider: local/ollama

Important

Using the exact same Codex Desktop installation, MCP server, API key, and MCP configuration:

  • GPT-5.6 + Brave Search MCP → works
  • Qwen3.6-35B-A3B + Brave Search MCP → unsupported call

This suggests that the MCP server and API key are working correctly, and that the issue is specific to MCP tool invocation with the local/custom model provider.

What steps can reproduce the bug?
  1. Configure the Brave Search MCP server in Codex.
  2. Start Codex Desktop.
  3. Use Qwen3.6-35B-A3B as the model.
  4. Ask the model to perform a web search using the Brave MCP tool.
  5. The MCP tool is visible in the available tools, but the invocation fails with:

unsupported call

What is the expected behavior?

MCP tools should be callable when using a compatible local/ollama provider, just as they are when using GPT-5.6.

Additional information

This appears to be related to the existing issues concerning MCP namespace tools and non-OpenAI/custom model providers.

However, I am reporting this separately because my reproduction is specifically on the Codex Desktop app, not the Codex CLI.

I hope this helps distinguish the Desktop reproduction from the CLI-related reports.

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start by reproducing the failure in the Codex Desktop app with the Brave Search MCP server, Ollama, and the named local model, then compare it with GPT-5.6. Review the existing issues about MCP namespace tools and non-OpenAI or custom model providers; done means compatible local-provider MCP calls succeed without the unsupported call error.

Written by the indexing model from the issue text.

Assessment

Domain
ai, desktop, devtools
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Quiet
Clarity
Needs clarification
Newbie friendliness
42/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.