OpenAI MCP Server

## Prerequisites - Node.js (version 18 or higher) - An OpenAI account with an API key (get one at [platform.openai.com/api-keys](https://platform.openai.com/api-keys)) - ChatGPT Desktop app (latest version) - ngrok account (free tier sufficient) or Cloudflare Tunnel for exposing local server via HTTPS - Basic familiarity with terminal commands ## Setup for ChatGPT Desktop ### Option 1: Using Developer Mode (Recommended) 1. Open ChatGPT Desktop and navigate to **Settings → Apps & Connectors → Advanced settings → Enable Developer Mode**. 2. Go to **Settings → Apps & Connectors → Create** to add a new connector. 3. Enter the MCP server URL (e.g., `https://your-ngrok-subdomain.ngrok-free.app/mcp` from ngrok). 4. Leave authentication blank unless custom auth is configured (default is none). 5. Save and test the connection—ChatGPT will now have access to OpenAI MCP tools. ### Option 2: Local Server Setup 1. Follow the start command to clone, install, and run the server locally (defaults to `http://localhost:3000`). 2. Install and authenticate ngrok: `ngrok authtoken YOUR_NGROK_TOKEN`. 3. Expose the server: `ngrok http 3000`. 4. Copy the HTTPS forwarding URL (e.g., `https://abc123.ngrok-free.app`) and append `/mcp` for the full endpoint. 5. Use this URL in ChatGPT Desktop's Developer Mode connector as described in Option 1. **Alternative with Cloudflare Tunnel:** ```bash # Install cloudflared brew install cloudflared # macOS, or equivalent for your OS # Run tunnel cloudflared tunnel --url http://localhost:3000 ``` Use the provided HTTPS URL. ## Configuration Examples ### Basic Setup Minimal config for quick start: ```bash export OPENAI_API_KEY=sk-your-key npm run dev ``` Server runs on `http://localhost:3000/mcp`. Tunnel with ngrok and connect to ChatGPT. ### Advanced Setup Customize models and options: ```bash export OPENAI_API_KEY=sk-your-key export OPENAI_BASE_URL=https://api.openai.com/v1 # Custom endpoint if needed export ALLOWED_MODELS=gpt-4o,gpt-4o-mini,gpt-4-turbo # Restrict models export PORT=8080 # Change port npm run dev ``` Update ngrok/tunnel to match the new port. ### Production Setup For reliable, secure use: 1. Use PM2 for process management: `npm i -g pm2`, then `pm2 start npm --name "openai-mcp" -- run start`. 2. Set up a reverse proxy (e.g., Nginx) with SSL certs from Let's Encrypt. 3. Use a paid ngrok/Cloudflare plan for static domains. 4. Add auth: `export MCP_AUTH_TOKEN=your-secret-token` and configure ChatGPT connector with Bearer token. ## Using with ChatGPT - **Available Capabilities**: Tools like `list_openai_models`, `openai_chat_completion`, `openai_embeddings` become available, allowing ChatGPT Desktop to invoke OpenAI models (e.g., GPT-4o, GPT-4o-mini) for tasks like generation, analysis, or embeddings. - **Example Prompts**: - "List all available OpenAI models using the MCP tool." - "Use GPT-4o-mini via OpenAI MCP to summarize this article: [paste text]." - "Generate code with GPT-4o: Write a Python function for [task]." - **Limitations/Considerations**: Incurs OpenAI API costs based on usage; subject to rate limits and quotas. Local server adds slight latency. Ensure HTTPS for ChatGPT compatibility—no HTTP. ## Troubleshooting - **Server fails to start**: Verify Node.js version (`node -v`), reinstall deps (`rm -rf node_modules && npm install`), check port conflicts. - **ChatGPT connection error**: Confirm HTTPS URL (HTTP not supported); restart ngrok/tunnel; check firewall blocks port 3000. - **API 401/Invalid Key**: Double-check `OPENAI_API_KEY` export (use `echo $OPENAI_API_KEY`); regenerate key if needed. - **Tool not appearing in ChatGPT**: Refresh Developer Mode connector; restart ChatGPT Desktop; verify `/mcp` endpoint responds (curl `https://your-url/mcp`). - **Rate limit errors**: Monitor OpenAI dashboard; add `export MAX_TOKENS=4000` or upgrade plan. - Logs: Check server console for details; enable debug with `export DEBUG=mcp*`.