Deployment

MCP servers communicate over stdio (standard input/output) or SSE (server-sent events). The transport determines where and how you deploy.

Local Development

For local testing, run your server with ts-node or tsx:

npx tsx server.ts

Or compile and run:

npm run build
node dist/server.js

Your server listens on stdio by default. Connect using the MCP Inspector or Claude Desktop.

stdio (Local & Desktop)

Stdio is the simplest transport — your server runs as a child process of the MCP client. No network setup, no ports.

Use when: running with Claude Desktop, VS Code Copilot, or local development.

// claude_desktop_config.json
{
  "mcpServers": {
    "my-server": {
      "command": "node",
      "args": ["dist/server.js"],
      "env": {
        "API_KEY": "sk-..."
      }
    }
  }
}

No deployment needed — just build and point the config to your script.

Vercel (Serverless)

For remote access, deploy your MCP server as a Vercel serverless function. The function listens for incoming SSE connections from MCP clients.

// api/mcp.ts
import { createServer, tool } from "mcpcraft-sdk"

const server = createServer({ name: "my-server" })

server.add(tool({
  name: "get_weather",
  description: "Fetches weather for a city",
  input: {
    city: { type: "string", description: "City name" }
  },
  run: async ({ city }) => {
    const res = await fetch(
      `https://api.open-meteo.com/v1/forecast?latitude=...&current_weather=true`
    )
    return await res.json()
  }
}))

export default server

// vercel.json
{
  "functions": {
    "api/mcp.ts": {
      "maxDuration": 60
    }
  }
}

vercel env add API_KEY
vercel env add DATABASE_URL

vercel deploy --prod

Docker

For full control over runtime and scaling, or for deploying to container platforms.

FROM node:20-alpine

WORKDIR /app
COPY package.json package-lock.json ./
RUN npm ci --omit=dev

COPY dist/ ./dist/

EXPOSE 3000
CMD ["node", "dist/server.js"]

docker build -t mcp-server .
docker run -e API_KEY=sk-... mcp-server

Railway

Railway provides easy Node.js deployments with automatic HTTPS and environment variable management.

1. Create a Dockerfile or use Railway's Node.js builder (auto-detected)
2. Set environment variables in the Railway dashboard
3. Deploy by connecting your GitHub repo or using the Railway CLI:

railway up

Railway assigns a *.railway.app URL. Use this as your SSE endpoint.

Render

Render supports Node.js web services with free SSL and auto-deploys from Git.

1. Create a Web Service on Render
2. Connect your repository
3. Set:
   • Build Command: npm run build
   • Start Command: node dist/server.js
4. Add environment variables in the Render dashboard
5. Deploy

Your server will be available at https://your-service.onrender.com.

VPS (Bare Metal)

For maximum control, run your server on any VPS (DigitalOcean, Hetzner, AWS EC2).

# Build
npm run build

# Run with process manager
npm install -g pm2
pm2 start dist/server.js --name mcp-server
pm2 save
pm2 startup

Use a reverse proxy (nginx, Caddy) if exposing over SSE:

server {
    listen 80;
    server_name mcp.example.com;

    location / {
        proxy_pass http://localhost:3000;
        proxy_http_version 1.1;
        proxy_set_header Upgrade $http_upgrade;
        proxy_set_header Connection "upgrade";
    }
}

Production Checklist