mcp -chatgpt-app

This MCP server connects the ChatGPT macOS ecosystem to the Model Context Protocol by routing a local ChatGPT integration through uv (the universal VM/runner) so you can manage it as an MCP server. The configuration uses uvx to run the mcp-server-chatgpt-app package/directory, allowing you to interact with the ChatGPT setup via MCP tooling like MCP Inspector and Chatwise. To enable the server, add it to your MCP configuration (as shown in the README) and start it through your preferred MCP launcher that supports the uvx-based command (for example, through the standard MCP workflow or your editor's integration).

You can use MCP Inspector to inspect and interact with this server using stdio transport, and you can integrate with Chatwise by adding a stdio-based tool that launches uvx mcp-server-chatgpt-app. Be mindful of the caveat noted in the documentation: Chatwise may not close the MCP server automatically, so you should manually clean up any lingering processes with a command like pkill -f 'mcp-server-chatgpt-app' if needed. For local development, you can run the server directly with uv, using the directory path to the ChatGPT MCP server implementation and the appropriate run command, as shown in the examples.

Prerequisites:

• macOS with the ChatGPT app installed
• uv installed (astral uv) and available on your PATH
• Access to the mcp-server-chatgpt-app code directory or package

Step-by-step installation:

1. Install uv if you haven't already:
   • Follow the instructions at https://docs.astral.sh/uv/getting-started/installation/
2. Ensure you have the ChatGPT macOS app and the Ask ChatGPT on Mac shortcuts prepared as per the prerequisites.
3. Clone or obtain the mcp-server-chatgpt-app code to a local directory, e.g. ~/Developer/mcp-server-chatgpt-app
4. Install any runtime requirements for the Python project if applicable (e.g., python3, pipenv, or venv setup) depending on how the server is implemented. If a virtual environment is used, activate it.
5. Verify the server entry point (e.g., server.py or equivalent) is reachable in the directory you cloned.
6. Update your MCP configuration to include the chatgpt server as shown in the README (see the example mcp.json snippet). For example, your .mcp.json should contain:

{
  "mcpServers": {
    "chatgpt": {
      "command": "uvx",
      "args": ["mcp-server-chatgpt-app"],
      "env": {},
      "disabled": false,
      "autoApprove": []
    }
  }
}

7. Start the MCP server using the configured command from your MCP toolchain, e.g. through the MCP UI or by invoking the uvx command directly in a shell if your setup allows:

uvx mcp-server-chatgpt-app

8. If you are running locally for development, you can start the server directly with uv in the directory containing the entry script, for example:

uv --directory $HOME/Developer/mcp-server-chatgpt-app/src/mcp_server_chatgpt run server.py

9. Confirm the server is alive and accessible via MCP Inspector or your chosen tooling. Troubleshoot issues by checking for lingering processes (pkill -f 'mcp-server-chatgpt-app') and ensuring the correct directory paths are used.

Tips and notes:

• The chatgpt MCP server relies on uvx to launch the mcp-server-chatgpt-app directory as an MCP server; ensure the path provided in the args matches your local setup.
• If you use Chatwise, be aware that it may not automatically close the MCP server when you exit, so you may need to manually clean up running processes with pkill -f 'mcp-server-chatgpt-app'.
• The MCP Inspector example for stdio transport uses uv and a specific directory/command layout; adapt the Command and Arguments as needed for your environment.
• If you modify the server path or directory structure, update the mcp.json configuration accordingly to reflect the correct path in the uvx invocation.
• Keep your prerequisites up to date: macOS ChatGPT app, uv utilities, and the shortcuts for ChatGPT on Mac.
• There is no npm package associated with this server in the provided configuration (npm_package is null).