App

mctx — The best way to Build an MCP Server

@mctx-ai/mcp is the best way to Build an MCP Server. Register tools, resources, and prompts — the framework handles protocol negotiation, input validation, error sanitization, and CORS. You write the business logic.

Quick Start

import { createServer, T } from "@mctx-ai/mcp";

const server = createServer({
  instructions: "A greeting server. Use the greet tool to say hello.",
});

function greet(mctx, req, res) {
  res.send(`Hello, ${req.name}! (user: ${mctx.userId || "anonymous"})`);
}

greet.description = "Greet someone by name";
greet.input = {
  name: T.string({ required: true, description: "Name to greet" }),
};

server.tool("greet", greet);

export default { fetch: server.fetch };

That's a working MCP server. Run it locally with npx mctx-dev index.js.

Installation

Scaffold a new project (recommended):

npx create-mctx-server my-app
cd my-app
npm install
npx mctx-dev index.js

Use the template repo:

github.com/mctx-ai/example-app — click "Use this template" on GitHub.

Add to an existing project:

npm install @mctx-ai/mcp

Run the dev server:

npx mctx-dev index.js

Hot reload is included. Changes to index.js restart the server automatically.

Features

• Zero runtime dependencies — ships nothing you don't need
• TypeScript-ready — full .d.ts type definitions included
• Hot reload dev server — mctx-dev watches your files and restarts on change
• Input validation — JSON Schema validation via the T type system
• Error sanitization — secrets and stack traces never leak to clients
• MCP protocol handled — capability negotiation, JSON-RPC 2.0, CORS — all automatic
• Cloudflare Workers — exports a standard fetch handler, deploys anywhere

API

Handler Signature

Every handler — tools, resources, and prompts — uses the same three-argument signature:

function myHandler(mctx, req, res) {
  res.send("result");
}

Tools

Tools are functions AI can call — like API endpoints.

function search(mctx, req, res) {
  const results = db.query(req.query, { limit: req.limit });
  res.send(results);
}

search.description = "Search the database";
search.input = {
  query: T.string({ required: true, description: "Search query" }),
  limit: T.number({ default: 10, description: "Max results" }),
};

server.tool("search", search);

For long-running tools, report progress with res.progress(current, total):

async function migrate(mctx, req, res) {
  for (let i = 0; i < req.tables.length; i++) {
    await copyTable(req.tables[i]);
    res.progress(i + 1, req.tables.length);

... [View full README on GitHub](https://github.com/mctx-ai/app#readme)