MCP Servers for OpenAI o1 and Flux AI Integration

## Prerequisites - Node.js >= 18 and npm - OpenAI API key (with access to o1 models like `o1-preview` or `o1-mini`) - fal.ai account and API key (for Flux image generation) - ngrok account (free tier sufficient for local tunneling) - ChatGPT Desktop app (latest version) - Git (for cloning the repository) ## 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. Start the MCP server locally (see Start Command above). Note the local URL: `http://localhost:3000`. 3. Expose the server publicly using ngrok: ```bash ngrok http 3000 ``` Copy the HTTPS URL (e.g., `https://abc123.ngrok.io`). 4. In ChatGPT Desktop, go to **Settings → Apps & Connectors → Create**. 5. Enter the MCP endpoint URL: `https://abc123.ngrok.io/mcp` (append `/mcp`). 6. Save the connector. It will appear in your apps list for use in chats. ### Option 2: Local Server Setup For persistent local use without ngrok: 1. Run the server: `npm start`. 2. Use Cloudflare Tunnel: Install `cloudflared`, then: ```bash cloudflared tunnel --url http://localhost:3000 ``` Use the provided HTTPS URL for the ChatGPT connector (as in Option 1). 3. Configure ChatGPT Desktop connector with the tunnel's `/mcp` endpoint. ## Configuration Examples ### Basic Setup Create a `.env` file in the server directory: ```env OPENAI_API_KEY=sk-your-openai-api-key FAL_KEY=your-fal-ai-key PORT=3000 ``` Run `npm start`. Supports default `o1-preview` for reasoning and `flux-schnell` for images. ### Advanced Setup Extend `.env` for custom models/parameters: ```env OPENAI_API_KEY=sk-your-openai-api-key FAL_KEY=your-fal-ai-key OPENAI_MODEL=o1-mini FLUX_MODEL=flux-pro MAX_TOKENS=4096 TEMPERATURE=0.7 ``` Restart server. Allows fine-tuned streaming chats and image variants. ### Production Setup For reliable use: 1. Use PM2: `npm install -g pm2`, then `pm2 start npm --name "mcp-o1-flux" -- start`. 2. Deploy to VPS (e.g., Render, Fly.io) with HTTPS via Let's Encrypt. 3. `.env`: ```env OPENAI_API_KEY=sk-prod-key FAL_KEY=prod-fal-key NODE_ENV=production ``` 4. Use domain like `https://yourdomain.com/mcp` in ChatGPT connector. ## Using with ChatGPT - **Available Tools**: `o1_reason` for advanced reasoning with streaming chat completions and parameter control (e.g., temperature, max tokens); `flux_image` for generating high-quality images from prompts. - **Example Prompts**: - Reasoning: "Use o1_reason to solve this puzzle: [describe problem]. Explain step-by-step." - Images: "Generate a Flux image of a futuristic cityscape at sunset." - Combined: "Reason with o1 about climate change impacts, then illustrate with a Flux image." - **Limitations/Considerations**: API usage incurs OpenAI/fal.ai costs; o1 models have longer latency; rate limits apply (monitor via provider dashboards); requires stable internet for tunneling. ## Troubleshooting - **Server not starting**: Check Node.js version (`node -v`) and `npm install`. Verify API keys. - **ngrok/Cloudflare errors**: Ensure account auth (`ngrok authtoken <token>`). Use HTTPS only. - **Connector fails in ChatGPT**: Confirm `/mcp` endpoint responds (curl `https://your-url/mcp`). Check browser console for CORS/auth issues. - **Tool not appearing**: Restart ChatGPT Desktop after adding connector. Ensure server logs show no errors. - **API errors**: Validate keys have quotas (o1 access via OpenAI playground); Flux prompts must be safe (no violations). - **Port conflict**: Change `PORT` in `.env` and restart. ```