Simple OpenAI Assistant MCP Server

## Prerequisites - Node.js (v18 or higher) - npm (v9 or higher) - OpenAI API key (get one from [platform.openai.com/api-keys](https://platform.openai.com/api-keys)) - ChatGPT Desktop app (latest version with Developer Mode support) - ngrok or Cloudflare Tunnel (for exposing local server to HTTPS) - Git (for cloning the repo) ## Setup for ChatGPT Desktop ### Option 1: Using Developer Mode (Recommended) 1. Open ChatGPT Desktop and go to **Settings → Apps & Connectors → Advanced settings → Enable Developer Mode**. 2. Navigate to **Settings → Apps & Connectors → Create** to add a new connector. 3. Enter the MCP server URL (e.g., `https://your-ngrok-url.ngrok.io/mcp`—see Option 2 for local exposure). 4. For authentication, add header: `Authorization: Bearer $OPENAI_API_KEY` or use query param if supported. 5. Save and test the connector—ChatGPT will now have access to OpenAI Assistant tools. ### Option 2: Local Server Setup 1. Follow the start command to clone, install, and run the server locally (`http://localhost:8787/mcp`). 2. Expose the endpoint via HTTPS (MCP requires secure connection): - **ngrok**: Install ngrok, then run `ngrok http 8787`. Copy the HTTPS URL (e.g., `https://abc123.ngrok.io`). - **Cloudflare Tunnel**: Install cloudflared, run `cloudflared tunnel --url http://localhost:8787`. Use the generated URL. 3. Use the full MCP URL in ChatGPT Desktop connector: `https://your-tunnel-url/mcp`. 4. Restart the server if needed and verify connectivity. ## Configuration Examples ### Basic Setup Minimal env vars for quick start: ```bash export OPENAI_API_KEY=sk-... npm start ``` Server runs on `http://localhost:8787/mcp`. Expose with ngrok and add to ChatGPT. ### Advanced Setup Add persistence and custom ports: ```bash export OPENAI_API_KEY=sk-... export PORT=3000 export DB_PATH=./my_threads.db # Custom SQLite path npm start ``` Access via `http://localhost:3000/mcp`. ### Production Setup Secure with env vars and HTTPS reverse proxy: ```bash export OPENAI_API_KEY=sk-... export NODE_ENV=production export DB_PATH=/secure/path/threads.db npm start ``` Use PM2 for process management: `npm install -g pm2; pm2 start npm --name "mcp-server" -- start`. Expose via domain with SSL (e.g., Caddy/Nginx). ## Using with ChatGPT - **Available Tools**: `create_assistant`, `run_assistant`, `list_threads`, `manage_conversations`—enabling ChatGPT to build and interact with custom OpenAI Assistants. - **Example Prompts**: - "Create a coding assistant with tools for Python debugging and run it on this code: [paste code]." - "List my saved threads and continue the last conversation about project planning." - "Build a math tutor assistant and solve: integrate x^2 from 0 to 1." - **Limitations/Considerations**: OpenAI API costs apply; local SQLite is single-user; streaming works best with stable connections. Threads persist locally—backup DB_PATH regularly. ## Troubleshooting - **Connection Refused**: Ensure server is running (`curl http://localhost:8787/mcp`) and tunnel is active. - **Auth Errors**: Verify `OPENAI_API_KEY` is valid and passed correctly in connector headers. - **SQLite Issues**: Check file permissions on `DB_PATH`; delete `.db` file to reset. - **No Tools Visible**: Restart ChatGPT Desktop after adding connector; check Developer Mode logs. - **Ngrok Disconnects**: Use paid ngrok for static domains or switch to Cloudflare Tunnel. - **Rate Limits**: Monitor OpenAI usage; add `export MAX_TOKENS=4096` for limits.