MCP Server Starter

A small remote MCP server you can deploy to Railway in one click. It has the boring parts done: Streamable HTTP transport, a /health check, bearer-token auth, structured JSON logs, env-based config, and one sample tool to replace with your own.

Endpoints

Related MCP server: MCP Server Template (Python)

Configuration

Connect a client

After deploy, open the service's Variables tab to copy MCP_AUTH_TOKEN. Your URL is https://<your-service>.up.railway.app/mcp.

Claude Code

claude mcp add --transport http starter https://<your-service>.up.railway.app/mcp \
  --header "Authorization: Bearer <MCP_AUTH_TOKEN>"

Cursor and other clients (mcp.json)

{
  "mcpServers": {
    "starter": {
      "url": "https://<your-service>.up.railway.app/mcp",
      "headers": { "Authorization": "Bearer <MCP_AUTH_TOKEN>" }
    }
  }
}

Add your own tool

Tools are registered in buildServer() in src/server.js:

server.registerTool(
  "add",
  { description: "Add two numbers", inputSchema: { a: z.number(), b: z.number() } },
  async ({ a, b }) => ({ content: [{ type: "text", text: String(a + b) }] }),
);

The server is stateless: each request gets a fresh MCP server instance. That lets you scale replicas without sticky sessions. If your tools need state, keep it in a database (for example, add Railway Postgres and read DATABASE_URL).

Run locally

npm install
cp .env.example .env   # optional
npm start              # http://localhost:3000
npm test               # health, auth, and an MCP client round-trip

Before you expose real tools

The bearer token is a minimum. For multi-user or public servers, put OAuth 2.1 in front of /mcp, as the MCP authorization spec describes. Keep tools least-privilege: don't ship shell-exec or filesystem-write tools on a public URL.

License

MIT © Intent Solutions