Google MCP Router

# MCP Router Deployment Plan ## 🎯 Overview This plan outlines how to deploy your MCP Router for testing with live AI agents, including hosting options, security considerations, and integration steps. ## 🏗️ Current Status - ✅ **MCP Router fully functional** (5 tools implemented) - ✅ **OAuth authentication working** (Google Calendar + Gmail) - ✅ **Local development complete** (localhost:8080) - ✅ **All dependencies resolved** (TypeScript, Fastify, Google APIs) ## 🚀 Deployment Options ### Option 1: Railway (Recommended for Testing) **Best for: Quick deployment, easy scaling, built-in monitoring** **Pros:** - Zero-config deployment from GitHub - Automatic HTTPS with custom domains - Built-in environment variable management - Free tier available - Easy database integration (Redis) **Steps:** 1. Push code to GitHub repository 2. Connect Railway to GitHub repo 3. Set environment variables in Railway dashboard 4. Deploy automatically 5. Get production URL (e.g., `https://your-app.railway.app`) **Cost:** Free tier (500 hours/month) + $5/month for production ### Option 2: Render **Best for: Simple deployment, good free tier** **Pros:** - Free tier with 750 hours/month - Automatic deployments from GitHub - Built-in SSL certificates - Easy environment management **Steps:** 1. Connect GitHub repository 2. Configure build command: `npm run build` 3. Set start command: `node dist/server.js` 4. Add environment variables 5. Deploy **Cost:** Free tier available ### Option 3: DigitalOcean App Platform **Best for: Production-ready, scalable** **Pros:** - Professional hosting platform - Auto-scaling capabilities - Integrated monitoring - Database hosting options **Steps:** 1. Create new app from GitHub 2. Configure Node.js runtime 3. Set environment variables 4. Deploy with custom domain **Cost:** $5/month minimum ### Option 4: Vercel (Serverless) **Best for: Edge deployment, global CDN** **Pros:** - Global edge deployment - Automatic HTTPS - Easy custom domains - Serverless scaling **Considerations:** - May need adjustments for long-running OAuth flows - Better for API endpoints than persistent connections ## 🔧 Pre-Deployment Setup ### 1. Environment Variables (Production) ```bash # Required for production NODE_ENV=production APP_PORT=8080 HOST=your-domain.com # Google OAuth (already configured) GOOGLE_CLIENT_ID=your-client-id GOOGLE_CLIENT_SECRET=your-client-secret GOOGLE_REDIRECT_URI=https://your-domain.com/oauth/google/callback # Security TOKEN_ENC_KEY=your-encryption-key # Optional REDIS_URL=redis://your-redis-url GMAIL_SENDER_EMAIL=your-email@gmail.com GMAIL_SENDER_NAME=Your Name ``` ### 2. Google OAuth Configuration Updates **Update Google Cloud Console:** 1. Go to [Google Cloud Console](https://console.cloud.google.com) 2. Navigate to APIs & Services > Credentials 3. Edit your OAuth 2.0 Client ID 4. Add production redirect URI: `https://your-domain.com/oauth/google/callback` 5. Save changes ### 3. Security Enhancements - [ ] Generate new encryption key for production - [ ] Set up Redis for idempotency storage - [ ] Configure CORS for your AI agent domain - [ ] Set up monitoring and logging ## 🤖 AI Agent Integration ### MCP Protocol Integration Your MCP Router exposes these endpoints for AI agents: #### **Base URL:** `https://your-domain.com` #### **Available Tools:** 1. **`time.resolve`** - Parse natural language time expressions 2. **`calendar.find_free_slots`** - Find available meeting times 3. **`calendar.create_event`** - Schedule meetings 4. **`calendar.cancel_event`** - Cancel meetings 5. **`email.send`** - Send templated emails #### **Authentication Flow:** 1. AI agent redirects user to: `GET /oauth/google/login` 2. User completes Google OAuth 3. Tokens stored securely in encrypted token store 4. AI agent can now use all tools with user's permissions ### Integration Examples #### **Claude Desktop Integration:** ```json { "mcpServers": { "google-mcp-router": { "command": "node", "args": ["/path/to/your/mcp-router/dist/server.js"], "env": { "NODE_ENV": "production" } } } } ``` #### **Custom AI Agent Integration:** ```python # Example Python integration import requests class MCPRouterClient: def __init__(self, base_url): self.base_url = base_url def resolve_time(self, input_text): response = requests.post( f"{self.base_url}/mcp/tools/time.resolve", json={"input": input_text} ) return response.json() def find_free_slots(self, start_time, end_time): response = requests.post( f"{self.base_url}/mcp/tools/calendar.find_free_slots", json={ "window_start_iso": start_time, "window_end_iso": end_time } ) return response.json() ``` ## 📋 Deployment Checklist ### Pre-Deployment - [ ] Choose hosting platform (Railway recommended) - [ ] Update Google OAuth redirect URIs - [ ] Generate production encryption key - [ ] Set up Redis instance (optional but recommended) - [ ] Test locally with production environment variables ### Deployment Steps - [ ] Push code to GitHub repository - [ ] Connect hosting platform to GitHub - [ ] Configure environment variables - [ ] Set up custom domain (optional) - [ ] Deploy and test OAuth flow - [ ] Test all MCP tools - [ ] Set up monitoring ### Post-Deployment - [ ] Verify OAuth authentication works - [ ] Test calendar operations - [ ] Test email sending - [ ] Configure AI agent integration - [ ] Set up error monitoring - [ ] Document API endpoints ## 🔒 Security Considerations ### Production Security - [ ] Use HTTPS only (automatic with most platforms) - [ ] Secure environment variables - [ ] Implement rate limiting - [ ] Set up CORS for specific domains - [ ] Monitor for suspicious activity - [ ] Regular security updates ### OAuth Security - [ ] Validate state parameters - [ ] Secure cookie handling - [ ] Token encryption at rest - [ ] Automatic token refresh - [ ] Secure token storage ## 📊 Monitoring & Maintenance ### Health Checks - [ ] `/health` endpoint monitoring - [ ] OAuth flow monitoring - [ ] API response time tracking - [ ] Error rate monitoring - [ ] Token refresh success rate ### Maintenance Tasks - [ ] Regular dependency updates - [ ] Security patch management - [ ] Performance optimization - [ ] Log rotation and cleanup - [ ] Backup procedures ## 🎯 Testing Strategy ### 1. Local Testing ```bash # Test OAuth flow curl -L http://localhost:8080/oauth/google/login # Test MCP tools curl -X POST http://localhost:8080/mcp/tools/time.resolve \ -H "Content-Type: application/json" \ -d '{"input": "next Monday at 2pm"}' ``` ### 2. Production Testing - [ ] OAuth authentication flow - [ ] All 5 MCP tools functionality - [ ] Error handling and edge cases - [ ] Performance under load - [ ] Security vulnerability scanning ### 3. AI Agent Integration Testing - [ ] Claude Desktop integration - [ ] Custom AI agent integration - [ ] End-to-end workflow testing - [ ] User experience validation ## 📈 Scaling Considerations ### Current Capacity - **Free Tier Limits:** - Railway: 500 hours/month - Render: 750 hours/month - Vercel: 100GB bandwidth/month ### Scaling Up - **Paid Plans:** $5-20/month for production use - **Database:** Add Redis for production idempotency - **CDN:** Use Cloudflare for global distribution - **Monitoring:** Add Datadog or New Relic ## 🚀 Next Steps 1. **Choose hosting platform** (Railway recommended) 2. **Set up GitHub repository** (if not already done) 3. **Configure production environment** 4. **Deploy and test OAuth flow** 5. **Integrate with AI agent** 6. **Monitor and optimize** ## 📞 Support & Troubleshooting ### Common Issues - **OAuth redirect mismatch:** Check Google Console settings - **CORS errors:** Configure allowed origins - **Token storage issues:** Verify encryption key - **Rate limiting:** Implement proper throttling ### Debug Tools - **Health endpoint:** `/health` - **Metrics endpoint:** `/metrics` - **Tool metadata:** `/mcp/tools` - **Server logs:** Platform-specific logging --- **Ready to deploy?** Choose your hosting platform and follow the deployment steps above! 🚀