OpenAI MCP Coding Assistant

## Prerequisites - Python 3.10 or higher - An OpenAI API key (get one from [platform.openai.com/api-keys](https://platform.openai.com/api-keys)) - Git for cloning the repository - ngrok or Cloudflare Tunnel for exposing local servers (HTTPS required for ChatGPT Desktop) - ChatGPT Desktop app (latest version) with Developer Mode enabled ## Setup for ChatGPT Desktop ### Option 1: Using Developer Mode (Recommended) 1. Open ChatGPT Desktop and go to **Settings → Apps & Connectors → Advanced settings**. 2. Enable **Developer Mode**. 3. Navigate to **Settings → Apps & Connectors → Create** and select **MCP Server**. 4. Enter the MCP server URL (e.g., `https://your-ngrok-url.ngrok.io/mcp`). 5. Add authentication if required (e.g., Bearer token via headers). 6. Save the connector. ChatGPT will now discover and use the server's tools. ### Option 2: Local Server Setup 1. Start the server locally (see Start Command above). It runs on `http://localhost:8787/mcp` by default. 2. Expose it via HTTPS: - **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 provided HTTPS URL. 3. Use the full MCP endpoint URL (e.g., `https://abc123.ngrok.io/mcp`) in ChatGPT's Developer Mode connector setup. ## Configuration Examples ### Basic Setup Use minimal environment variables for quick start: ```bash export OPENAI_API_KEY=sk-your-openai-key python main.py ``` Connect ChatGPT to `http://localhost:8787/mcp` (local) or ngrok URL. ### Advanced Setup Enable multi-LLM and custom options: ```bash export OPENAI_API_KEY=sk-your-openai-key export ANTHROPIC_API_KEY=your-anthropic-key # For additional LLM support export MAX_COST_PER_SESSION=10.00 # USD limit export TOOLS_ENABLED=file,search,shell # Customize tools python main.py --port 8787 --debug ``` Supports other LLMs like Anthropic via env vars. ### Production Setup For secure, persistent use: ```bash # Use .env file echo "OPENAI_API_KEY=sk-prod-key" > .env echo "MCP_HOST=0.0.0.0" >> .env echo "LOG_LEVEL=info" >> .env # Run with PM2 or systemd for daemon pm2 start main.py --name mcp-coding-assistant ``` Deploy on VPS with Cloudflare Tunnel or reverse proxy (e.g., Nginx) for HTTPS. ## Using with ChatGPT Once connected, ChatGPT gains access to: - **File operations**: Read/write/edit files, list directories. - **Search**: Code search, web search integration. - **Shell execution**: Run commands, manage processes. - **Real-time visualization**: Tool calls shown inline with costs. - **Cost management**: Tracks and limits API usage. **Example Prompts**: - "Create a new Python file called `hello.py` in my project folder and write a simple web server." - "Search my codebase for 'def process_data' and explain the function." - "Run `ls -la` in the current directory and summarize the output." - "Optimize this SQL query: [paste query] and save the improved version." **Limitations/Considerations**: - Local servers need HTTPS tunneling. - Shell execution is powerful—use sandboxed environments. - Costs accrue from LLM calls; monitor via built-in tracking. - Tools are read-eval only by default; enable writes cautiously. ## Troubleshooting - **Connection failed**: Ensure HTTPS URL ends with `/mcp` and server is running (`curl http://localhost:8787/mcp`). - **API key errors**: Verify `OPENAI_API_KEY` is valid and has credits. - **ngrok issues**: Free ngrok changes URLs on restart—use paid for static domains. - **Tools not appearing**: Restart ChatGPT Desktop after connector setup; check server logs for errors. - **Port conflicts**: Change `MCP_PORT` env var. - **High latency**: Use production setup; avoid local for heavy use. - Logs: Run with `--debug` for verbose output.