Go DALL-E Image Generation MCP Server

## Prerequisites - **Go 1.21+**: Download and install from [https://go.dev/doc/install](https://go.dev/doc/install). - **OpenAI API Key**: Create an account at [https://platform.openai.com/api-keys](https://platform.openai.com/api-keys) and generate a key with DALL-E access (requires billing enabled for paid usage). - **ngrok or Cloudflare Tunnel**: For exposing local server to HTTPS (free tier sufficient). - **ChatGPT Desktop app**: Latest version with Developer Mode support. - Git for cloning the repository. ## Setup for ChatGPT Desktop ### Option 1: Using Developer Mode (Recommended) 1. Open ChatGPT Desktop app. 2. Go to **Settings → Apps & Connectors → Advanced settings → Enable Developer Mode**. 3. Start the MCP server locally (see Start Command above; it runs on `http://localhost:8080`). 4. Install and run ngrok: `ngrok http 8080` (sign up for free authtoken if needed). Copy the HTTPS URL (e.g., `https://abc123.ngrok-free.app`). 5. In ChatGPT: **Settings → Apps & Connectors → Create**. 6. Enter **Connector URL**: `https://abc123.ngrok-free.app/mcp` (append `/mcp`). 7. **Authentication**: None (uses env vars on server). 8. Save and test the connector. ### Option 2: Local Server Setup 1. Follow the Start Command to run the server on `localhost:8080`. 2. Use Cloudflare Tunnel: - Install: `brew install cloudflared` (macOS) or equivalent. - Run: `cloudflared tunnel --url http://localhost:8080`. - Use the provided HTTPS URL + `/mcp` in ChatGPT connector. 3. Verify server logs show "Server listening on :8080" and test endpoint: `curl http://localhost:8080/mcp/health`. ## Configuration Examples ### Basic Setup ```bash export OPENAI_API_KEY="sk-..." go run . ``` - Saves images to `./images` with 1024x1024 default. ### Advanced Setup ```bash export OPENAI_API_KEY="sk-..." export IMAGE_SAVE_PATH="/path/to/custom/images" export DEFAULT_SIZE="1792x1024" # DALL-E 3 HD landscape export DALLE_MODEL="dall-e-3" export PORT=9090 # Custom port go run . ``` Update ngrok/Cloudflare to match port. ### Production Setup - Use systemd/PM2 equivalent for Go (e.g., `nohup go run . &`). - Secure with Cloudflare Zero Trust Tunnel (requires account). - Set `IMAGE_SAVE_PATH` to persistent volume. - Monitor OpenAI usage dashboard for costs. - HTTPS only via tunnel; avoid direct localhost. ## Using with ChatGPT - **Available Tools**: `generate_image` (text-to-image), `generate_variations` (image edits/variations). - **Example Prompts**: - "Use DALL-E to generate a futuristic cityscape at night." - "Create a variation of [describe image] with more vibrant colors." - "Generate a 1024x1024 portrait of a dragon in watercolor style." - Images auto-save locally and return URLs/paths in responses. - **Limitations/Considerations**: - DALL-E 3/2 API costs apply (~$0.04–0.12/image). - Rate limits: 50 images/min (adjustable via OpenAI tiers). - No internet images; text prompts only. - Custom sizes: 1024x1024, 1792x1024, etc. (DALL-E 3 supports HD). - Server must stay running; reconnect if ngrok URL expires. ## Troubleshooting - **"Invalid API Key"**: Verify `OPENAI_API_KEY` env var and DALL-E permissions. - **Connector fails**: Ensure HTTPS URL with `/mcp`, check server logs for errors, test `curl https://your-url/mcp`. - **Ngrok connection refused**: Server not running or wrong port; restart ngrok after server. - **No images generated**: Check OpenAI quota/billing; prompt too vague (add details like style/size). - **Port conflict**: Set `PORT=9090` env and update tunnel. - **Logs**: Server prints to stdout; tail with `go run . 2>&1 | tee server.log`. - **Go errors**: Run `go mod tidy`; ensure Go 1.21+ (`go version`).