Dev

@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.

MCP (Model Context Protocol) is the open standard AI assistants use to call external tools. Claude, ChatGPT, Cursor, and other clients speak it. Build an MCP server once, and it's accessible from every MCP-compatible AI.

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);
  }
  res.send(`Migrated ${req.tables.length} tables`);
}

migrate.description = "Migrate database tables";
migrate.input = {
  tables: T.array({ required: true, items: T.string() }),
};

server.tool("migrate", migrate);

Resources

Resources are read-only data AI can pull for context. Use static URIs or URI templates.

// Static resource
function readme(mctx, req, res) {
  res.send("# My Project\nWelcome to the docs.");
}

readme.mimeType = "text/markdown";
server.resource("docs://readme", readme);

// Dynamic template — {userId} is extracted and available on req
function getUser(mctx, req, res) {
  res.send(JSON.stringify(db.findUser(req.userId)));
}

getUser.description = "Fetch a user by ID";
getUser.mimeType = "application/json";
server.resource("user://{userId}", getUser);

Prompts

Prompts are reusable message templates for initializing AI conversations.

function codeReview(mctx, req, res) {
  res.send(`Review this ${req.language} code for bugs:\n\n${req.code}`);
}

codeReview.description = "Review code for issues";
codeReview.input = {
  code: T.string({ required: true }),
  language: T.string({ description: "Programming language" }),
};

server.prompt("code-review", codeReview);

For multi-message prompts with images or embedded resources, use conversation():

import { conversation } from "@mctx-ai/mcp";

function debug(mctx, req, res) {
  res.send(
    conversation(({ user, ai }) => [
      user.say("I hit this error:"),
      user.say(req.error),
      user.attach(req.screenshot, "image/png"),
      ai.say("I'll analyze the error and screenshot together."),
    ]),
  );
}

debug.description = "Debug with error + screenshot";
debug.input = {
  error: T.string({ required: true }),
  screenshot: T.string({ required: true, description: "Base64 image data" }),
};

server.prompt("debug", debug);

LLM Sampling

Use res.ask() to request an LLM completion from the client (LLM-in-the-loop):

async function summarize(mctx, req, res) {
  const content = await fetchPage(req.url);
  const summary = res.ask ? await res.ask(`Summarize:\n\n${content}`) : content;
  res.send(summary);
}

res.ask is null when the client does not support sampling — always check before calling.

Type System

T builds JSON Schema definitions for tool and prompt inputs.

All types accept required, description, and default.

Logging

import { log } from "@mctx-ai/mcp";

log.info("Server started");
log.warning("Rate limit approaching");
log.error("Connection failed");

Levels follow RFC 5424: debug, info, notice, warning, error, critical, alert, emergency.

Deploy

Push to GitHub and connect your repo at mctx.ai. Your server goes live.

Full deployment guide at docs.mctx.ai.

Contributing

See CONTRIBUTING.md and GitHub Issues.

Links

• Documentation
• Example Server
• npm: @mctx-ai/mcp