MCP OpenAI Server

You can run OpenAI models inside Claude through an MCP server, giving you seamless access to multiple OpenAI models directly from Claude’s chat interface. This setup lets you switch models on the fly and send messages to OpenAI without leaving your conversation flow.

How to use

Start a chat with Claude. When you want to use OpenAI’s models, tell Claude to switch to an OpenAI model for your current task. You can reference models such as gpt-4o, gpt-4o-mini, o1-preview, or o1-mini. For example, you might say Can you ask o1 what it thinks about this problem? or What does gpt-4o think about this?.

Tools available through the OpenAI MCP server include a chat endpoint that sends messages to OpenAI’s chat completion API. You can specify the model to use, with gpt-4o as the default when you don’t specify one.

How to install

Prerequisites you need before installing the MCP server:

• Node.js 18 or newer (this includes npm and npx)

• Claude Desktop app

• OpenAI API key

Install and configure the MCP server by adding the following entry to your Claude Desktop configuration file. On macOS, the file is located at ~/Library/Application Support/Claude/claude_desktop_config.json.

{
  "mcpServers": {
    "mcp-openai": {
      "command": "npx",
      "args": ["-y", "@mzxrai/mcp-openai@latest"],
      "env": {
        "OPENAI_API_KEY": "your-api-key-here (get one from https://platform.openai.com/api-keys)"
      }
    }
  }
}

Additional setup notes

This configuration enables Claude Desktop to start the OpenAI MCP server whenever you need it. The server supports multiple OpenAI models and includes basic error handling to help you troubleshoot common issues quickly.

Configuration details and notes

The MCP server is designed to run locally via a simple command using npx. The key configuration shown creates a standard runtime where Claude launches the MCP server on demand, and you provide your OpenAI API key for authentication.

Troubleshooting

If you run into issues, view MCP logs to investigate problems. On macOS you can tail the logs with: tail -n 20 -f ~/Library/Logs/Claude/mcp*.log.

Available tools

openai_chat

Sends messages to OpenAI's chat completion API. You specify the messages array and the model to use; defaults to gpt-4o if not provided.