Summary

One model asking another, nothing more.

This is the smallest useful shape of a cross-vendor bridge: no routing, no cost controls, no conversation state — a prompt goes to OpenAI and the answer comes back. Usage bills to your own OpenAI key, so treat it as a second meter running alongside the first.

What it is

A minimal MCP server that lets Claude query OpenAI models directly over the Model Context Protocol. It is a single-purpose bridge: the prompt goes out through your OpenAI key and the response comes back into the conversation.

What you get

A second model's answer available inside the first model's conversation, for comparison or a second opinion
A small enough codebase to read end to end before you trust it with a key — install is `pip install -e .` from a clone
A pytest check that exercises a real API call, so you can confirm the key and the wiring before connecting a client

Requirements

An OpenAI API key in `OPENAI_API_KEY`, and `PYTHONPATH` pointed at the checkout, both set in the client's server entry. It launches as `python -m src.mcp_server_openai.server`. The pyproject names the package `mcp-server-openai` at 0.1.0, with the console script `mcp-server-openai`. MIT licensed.

Setup effort

Build from source — clone the repository and build it, then point your client at the binary