Node.js

The @modelfetch/node package lets you run MCP servers as traditional Node.js HTTP servers.

Installation

Terminal

npm install @modelfetch/node

Usage

Start The Server

src/index.ts

import handle from "@modelfetch/node";
import server from "./server"; // Import your McpServer

// Run as a Node.js HTTP server
handle(server);

Log The Endpoint

src/index.ts

import handle, { getEndpoint } from "@modelfetch/node";
import server from "./server"; // Import your McpServer

// Run as a Node.js HTTP server
handle(server, (address) => {
  // Log the endpoint when the server starts listening
  console.log(`MCP server is available at ${getEndpoint(address)}`);
});

Specify Custom Port

src/index.ts

import handle, { getEndpoint } from "@modelfetch/node";
import server from "./server"; // Import your McpServer

// Run as a Node.js HTTP server
handle(
  server,
  (address) => {
    console.log(`MCP server is available at ${getEndpoint(address)}`);
  },
  // Customize server options
  { port: 8080 },
);

API Reference

`handle(server, callback?, options?)`

Starts the MCP server

server: Required McpServer instance from @modelcontextprotocol/sdk
callback: Optional listening callback that receives the server AddressInfo
options: Optional configuration object
- port: Custom port number (default: 3000)
- hostname: Custom hostname (default: "localhost")
- createServer: Custom server factory from node:http, node:https, or node:http2
- serverOptions: Custom server options from node:http, node:https, or node:http2

`getEndpoint(address)`

Gets the MCP server endpoint from the server AddressInfo

address: Required server AddressInfo from the listening callback

On this page