Summary
A second opinion from a different model, in the session you are already working in.
Because it runs on the Responses API, the thread persists across calls — so this is a conversation with ChatGPT rather than a series of disconnected prompts, and the web-search variant covers the questions your local model's knowledge cannot.
What it is
A bridge to OpenAI's Responses API: send a prompt to ChatGPT from another assistant, optionally with web search turned on, and keep the thread going across turns.
What you get
- A ChatGPT call with the model version, temperature and other parameters set by you
- A second tool that runs the same call with web search enabled, for questions that need current information
- Conversation state managed automatically by the Responses API, so a back-and-forth continues rather than restarting
- Long running discussions between two assistants, since each turn carries the thread forward
Requirements
Your own OpenAI API key in OPENAI_API_KEY, with DEFAULT_MODEL setting which model is used. Python 3.10 or higher and uv for the environment.
Setup effort
One command plus a key — npx -y @smithery/cli install @billster45/mcp-chatgpt-responses --client claude, then supply credentials