# billster45/mcp-chatgpt-responses [Health: Active]

**Category:** 🛠️ Other Tools and Integrations  
**Repository:** https://github.com/billster45/mcp-chatgpt-responses  
**GitHub Stars:** 22  
**npm Downloads (last month):** 33418  
**Views:** 3  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/billster45-mcp-chatgpt-responses

## Description
MCP server for Claude to talk to ChatGPT and use its web search capability.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "mcp-chatgpt-responses": {
    "command": "npx",
    "args": ["-y","@smithery/cli"],
    "env": {
      "OPENAI_API_KEY": ""
    }
  }
}
```

**Requires environment variables:** `OPENAI_API_KEY` — the values above are empty placeholders; fill in real credentials before running (see the repository for what each one is for).

## Documentation

## What billster45/mcp-chatgpt-responses does

The billster45/mcp-chatgpt-responses MCP server connects Claude Desktop with OpenAI's ChatGPT API. It exposes tools for sending prompts, selecting model and generation settings, and optionally enabling web search. The server is intended for workflows where Claude asks ChatGPT to answer a question, compares responses between the two models, or delegates requests that benefit from current web information.

An OpenAI API key is supplied by the user through the server configuration. Claude Desktop then invokes the MCP tools rather than calling OpenAI directly from the conversation interface.

## How it works

Requests use OpenAI's Responses API. The API returns a response identifier that can be passed into a later request, allowing OpenAI to retain the conversation state across turns. This supports longer exchanges between Claude and ChatGPT without the client having to rebuild the complete message history for every call.

Two request paths are available:

- `ask_chatgpt` sends a prompt to ChatGPT with optional model, temperature, output-token, and response-state parameters.
- `ask_chatgpt_with_web_search` sends a prompt with web search enabled for information that may have changed recently.

The web-search tool can also be called repeatedly in an agentic workflow, such as checking weather and then using the result to refine a plan.

## Setup and configuration

The documented prerequisites are Python 3.10 or newer, Claude Desktop, an OpenAI API key, and `uv`. Installation involves cloning the repository, creating a virtual environment with `uv venv`, activating it, and installing the dependencies from `requirements.txt` with `uv pip install`.

Claude Desktop is configured with an MCP server entry that runs `chatgpt_server.py` through `uv`. The example configuration supplies `OPENAI_API_KEY` and can set `DEFAULT_MODEL`, `DEFAULT_TEMPERATURE`, and `MAX_TOKENS`. The repository also documents installation through Smithery, but the local configuration requires the path to the cloned repository to be adjusted before use. Claude Desktop must be restarted after configuration changes.

## Tools and capabilities

The billster45/mcp-chatgpt-responses MCP server provides the following tools:

- `ask_chatgpt`: sends a prompt and returns a ChatGPT response.
- `ask_chatgpt_with_web_search`: asks ChatGPT to use web search while answering.

Both tools accept model and temperature options, a maximum output-token value, and a response identifier for continued conversation state. The server supports configurable default model, temperature, and token settings through environment variables.

## Limitations and notes

The supplied documentation specifically describes use with Claude Desktop; it does not document compatibility with other MCP clients. Web search is available through the dedicated web-search tool, not necessarily for every ordinary ChatGPT request. An OpenAI API key is required, and requests depend on the availability and behavior of OpenAI's Responses API. The setup example contains a local repository path that must be replaced with the actual checkout location; it is not a ready-to-run universal command.

The billster45/mcp-chatgpt-responses MCP server is released under the MIT License.

## Getting started with this billster45/mcp-chatgpt-responses MCP server
Always refer to the official documentation for the most accurate and up-to-date information.

_Full upstream README: https://allmcps.com/mcp/billster45-mcp-chatgpt-responses/readme_

