Customer Registration MCP Server

# Deployment Guide This project can be deployed in two modes: ## 1. Full Mode (with MCP Integration) - Recommended for VPS/Cloud For platforms that support long-running Node.js processes (AWS EC2, DigitalOcean, Heroku, Railway, Render, etc.) ### Requirements - Node.js 18+ - Ability to run long-running processes - Support for stdio communication ### Environment Variables ```env OPENAI_API_KEY=sk-your-key CUSTOMER_API_HOST=https://your-api.com CUSTOMER_API_TOKEN=your-token AGENT_TONE=Professional, helpful, and efficient CHATBOT_PORT=3000 NODE_ENV=production ``` ### Deployment Steps ```bash # 1. Build the project yarn install yarn build # 2. Start the chatbot server yarn chatbot:start # Or with PM2 pm2 start build/chatbotServer.js --name chatbot # Or with systemd sudo systemctl start chatbot ``` ### Features Available - ✅ Full MCP integration - ✅ Real customer creation via API - ✅ Conversation history - ✅ All chatbot features --- ## 2. Serverless Mode (Vercel) - Limited Functionality **⚠️ Important Limitations:** - MCP server **cannot** run on Vercel (requires stdio/long-running process) - Customer creation is **simulated** (not actually created in API) - No conversation history between requests - OpenAI responds but doesn't execute MCP actions ### Vercel Deployment 1. **Push to GitHub:** ```bash git push origin main ``` 2. **Connect to Vercel:** - Go to [vercel.com](https://vercel.com) - Import your GitHub repository - Configure environment variables in Vercel dashboard 3. **Required Environment Variables in Vercel:** ``` OPENAI_API_KEY=sk-your-key AGENT_TONE=Professional, helpful, and efficient ``` 4. **Deploy:** - Vercel will auto-deploy on push - Visit your deployment URL ### What Works on Vercel - ✅ Chat UI (beautiful interface) - ✅ OpenAI conversations - ✅ Data collection and formatting - ❌ Actual customer creation (MCP not available) - ❌ Conversation persistence --- ## 3. Recommended Deployment Options ### Option A: Railway.app (Easy + Full Features) 1. Push code to GitHub 2. Connect Railway to your repo 3. Add environment variables 4. Deploy - Railway will run `yarn chatbot:start` 5. Get a public URL automatically **Pros:** Full MCP support, easy setup, automatic HTTPS **Cons:** Paid service (free tier available) ### Option B: DigitalOcean App Platform 1. Connect your GitHub repo 2. Set build command: `yarn build` 3. Set run command: `yarn chatbot:start` 4. Add environment variables 5. Deploy **Pros:** Full features, reliable, good pricing **Cons:** Requires payment ### Option C: AWS EC2 + PM2 1. Launch EC2 instance 2. Install Node.js and PM2 3. Clone repo and build 4. Run with PM2: `pm2 start build/chatbotServer.js` 5. Configure nginx as reverse proxy **Pros:** Full control, scalable, production-ready **Cons:** More setup required, manual SSL configuration ### Option D: Render.com 1. Connect GitHub repo 2. Choose "Web Service" 3. Build: `yarn build` 4. Start: `yarn chatbot:start` 5. Add environment variables **Pros:** Easy setup, free tier, full MCP support **Cons:** Free tier sleeps after inactivity --- ## Environment Variables Reference ### Required for All Deployments ```env OPENAI_API_KEY=sk-... # OpenAI API key for GPT-4o-mini ``` ### Required for Full Mode (MCP) ```env CUSTOMER_API_HOST=https://... # Your customer API endpoint CUSTOMER_API_TOKEN=... # Bearer token for API ``` ### Optional ```env AGENT_TONE=... # Chatbot personality (default: Professional, helpful, and efficient) CHATBOT_PORT=3000 # Server port (default: 3000) NODE_ENV=production # Environment mode ``` --- ## Testing Your Deployment ### Health Check ```bash curl https://your-domain.com/health ``` ### Chat Test ```bash curl -X POST https://your-domain.com/api/chat \ -H "Content-Type: application/json" \ -d '{"message": "Hello!"}' ``` --- ## Performance Considerations ### Full Mode (VPS) - Single instance can handle 100-500 concurrent users - MCP process adds ~50MB memory overhead - Consider horizontal scaling for > 1000 users ### Serverless Mode (Vercel) - Auto-scales to millions of requests - Cold starts ~200-500ms - No MCP = faster responses - Consider for demo/preview only --- ## Troubleshooting ### "MCP request timeout" - Check MCP server is running: `ps aux | grep chatbotServer` - Verify environment variables are set - Check logs: `tail -f /var/log/chatbot.log` ### "OpenAI API error" - Verify `OPENAI_API_KEY` is correct - Check API credits/quota - Ensure `gpt-4o-mini` access ### "Customer API error" - Verify `CUSTOMER_API_HOST` and `CUSTOMER_API_TOKEN` - Test API directly with curl - Check API is reachable from deployment --- ## Monitoring ### Logs ```bash # PM2 pm2 logs chatbot # systemd journalctl -u chatbot -f # Vercel vercel logs ``` ### Metrics - Response times - Error rates - OpenAI API usage - Memory/CPU usage --- ## Security Checklist - [ ] Environment variables not committed to git - [ ] HTTPS enabled - [ ] API keys rotated regularly - [ ] CORS configured properly - [ ] Rate limiting enabled (production) - [ ] Error messages don't expose secrets - [ ] Dependencies updated --- ## Cost Estimates ### OpenAI Usage (GPT-4o-mini) - ~$0.00015 per request (average) - 1000 requests = ~$0.15 ### Hosting - **Vercel**: Free (with limits) - **Railway**: $5-20/month - **Render**: Free tier available, $7+/month - **DigitalOcean**: $6-12/month - **AWS EC2**: $5-30/month --- ## Support For deployment issues: 1. Check logs first 2. Verify environment variables 3. Test locally first 4. Check platform-specific documentation --- ## License MIT