mcp -openai

This MCP server enables querying OpenAI models directly from Claude using the MCP protocol. It exposes an OpenAI-based endpoint through the MCP server named openai-server, which runs as a Python module (src.mcp_server_openai.server). To use it, configure your Claude Desktop MCP settings to point to this server entry, providing your OpenAI API key via the OPENAI_API_KEY environment variable. You can then send prompts to specified OpenAI models through the MCP interface, and receive responses back into Claude just like other MCP-connected services. Ensure your PYTHONPATH points to the local mcp-server-openai project so the module can be discovered, and supply a valid API key to authorize calls to OpenAI.

Prerequisites:

• Python 3.8+ installed on your system
• Git
• Access to install Python packages (pip)

Steps:

1. Clone the repository: git clone https://github.com/pierrebrunelle/mcp-server-openai
2. Install in editable mode (from the repository root): pip install -e .
3. Set up your environment variables when running the server:
   • Ensure OPENAI_API_KEY is set to your OpenAI API key
   • Set PYTHONPATH to the path of your local mcp-server-openai project (as shown in the config example)
4. Start or configure the MCP server entry as shown in the mcp_config example, using the Python module invocation: python -m src.mcp_server_openai.server
5. In Claude Desktop, add or update the mcpServers entry to point to the above server configuration.

Notes:

• If you modify code, reinstall or use pip install -e . to pick up changes.
• Ensure network access to OpenAI APIs is allowed from the host running the server.

Tips and considerations:

• Securely manage your OpenAI API key; avoid committing it to version control.
• The PYTHONPATH must point to the project root so the module src.mcp_server_openai.server is importable.
• If you encounter connection or authentication errors, verify the OPENAI_API_KEY environment variable is correctly set in the runtime environment.
• Check network proxies or firewalls that may block outbound HTTPS to OpenAI endpoints.
• For testing, you can run pytest as shown in the README to validate the OpenAI call flow, but ensure you provide a valid API key to avoid test failures.
• This MCP server is designed to be used from Claude Desktop via the MCP protocol; for debugging, you can run the Python server directly and send test prompts to ensure it forwards requests correctly.