Mcpcraft

npm version License: MIT

Build MCP servers in ~10 lines instead of ~100.

A lightweight TypeScript SDK for building Model Context Protocol (MCP) servers — zero boilerplate, full type safety, automatic schema generation.

Why mcpcraft-sdk?

The official @modelcontextprotocol/sdk requires ~100 lines of boilerplate per server. mcpcraft-sdk distills that down to ~10 lines — with automatic schema generation, type inference, and input validation.

import { createServer, tool } from "mcpcraft-sdk"

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

server.add(tool({
  name: "greet",
  description: "Greets a user by name",
  input: {
    name: { type: "string", description: "The user's name" }
  },
  run: async ({ name }) => {
    return { message: `Hello, ${name}!` }
  }
}))

server.start()

That's a complete, working MCP server. ✅

What is MCP?

The Model Context Protocol (MCP) is an open standard that lets AI apps (Claude, Cursor, VS Code Copilot) call functions and read data from your server — the same way HTTP lets browsers talk to APIs. MCP is to AI what HTTP is to the web.

Install

npm install mcpcraft-sdk
# or
pnpm add mcpcraft-sdk
# or
yarn add mcpcraft-sdk

Quick Start

1. Create your server

import { createServer, tool } from "mcpcraft-sdk"

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

server.add(tool({
  name: "get_weather",
  description: "Get the current weather for a location",
  input: {
    location: { type: "string", description: "City name" }
  },
  run: async ({ location }) => {
    return { temp: 22, condition: "Sunny", city: location }
  }
}))

server.start()

2. Run it

npx ts-node server.ts

3. Connect it

Add to your MCP client config:

{
  "mcpServers": {
    "my-first-server": {
      "command": "node",
      "args": ["dist/server.js"]
    }
  }
}

Architecture

AI Agent (Claude, Cursor)
  │
  ▼  JSON-RPC over stdio/SSE
MCP Client
  │
  ▼
Your MCP Server (mcpcraft-sdk)
  │
  ├── Tools (executable functions)
  └── Resources (read-only data)

API Reference

createServer(options)

Creates a new MCP server instance with automatic transport setup.

Param	Type	Required	Default	Description
name	string	Yes	—	Server name (shown to MCP clients)
version	string	No	"0.1.0"	SemVer version
description	string	No	—	Human-readable description

Methods:

• server.add(item) — Register a tool or resource (chainable)
• server.start() — Launch the server on stdio transport
• server.server — Access the underlying raw SDK Server instance

tool(options)

Defines an executable function that LLMs can invoke.

Param	Type	Required	Description
name	string	Yes	Alphanumeric identifier
description	string	Yes	Explains the tool to the LLM
input	InputSchema	No	Parameter definitions
run	(args) => any	Yes	Execution handler (typed args)

Input Schema:

input: {
  query:   { type: "string",  description: "Search query" },
  limit:   { type: "number",  description: "Max results", required: false },
  verbose: { type: "boolean", description: "Detailed output", required: false }
}

resource(options)

Defines a read-only data source identified by URI.

Param	Type	Required	Description
name	string	Yes	Resource identifier
description	string	Yes	Describes the data to the LLM
uri	string	Yes	URI or URI template (users://{id}/profile)
mimeType	string	No	MIME type (default: application/json)
fetch	(params) => any	Yes	Returns the resource data

Documentation

Full documentation is available at mcpcraft.vercel.app:

• Getting Started
• Quick Start
• Deployment
• API Reference
• Examples

License

MIT