Quick Start

Create your first MCP server in under 5 minutes.

1. Install

Create a new project and install mcpcraft-sdk:

mkdir my-mcp-server
cd my-mcp-server
npm init -y
npm install mcpcraft-sdk zod

2. Write your server

Create server.ts:

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

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

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()

input: { name: { type: "string" } } looks like a simple object, but behind the scenes mcpcraft-sdk generates a JSON Schema for the MCP protocol and a Zod validator for runtime checks — both automatically.

3. Run it

npx ts-node server.ts

The server starts and listens on stdio. It won't print anything — it's waiting for an MCP client to connect.

4. Test it

Install the MCP Inspector to test your server locally:

npx @modelcontextprotocol/inspector node -- dist/server.js

Or, if you're using Claude Desktop, add this to your claude_desktop_config.json:

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

What just happened?

Let's walk through the code line by line:

Code	What it does
`createServer({ name })`	Initializes an MCP server instance with stdio transport
`tool({ name, description, input, run })`	Defines a callable function — name tells the LLM what it does, description helps the LLM decide when to use it
`input: { name: { type: "string" } }`	Declares parameters → mcpcraft-sdk auto-generates JSON Schema + Zod validator
`run: async ({ name }) => { ... }`	The handler that executes when the LLM calls this tool — receives typed, validated arguments
`server.add(tool(...))`	Registers the tool with the MCP protocol so the client can discover it via `tools/list`
`server.start()`	Launches the stdio transport and begins listening for JSON-RPC messages

What the LLM sees

When a user asks "Greet me as Alice", this is what happens behind the scenes:

Claude sees your greet tool in the MCP tool list
It reads the description: "Greets a user by name"
It sees it needs a name parameter (string)
It calls greet({ name: "Alice" })
Your handler runs and returns { message: "Hello, Alice!" }
Claude reads the result and responds: "Hello, Alice! How can I help you today?"

Quick Start

Quick Start

1. Install

2. Write your server

3. Run it

4. Test it

What just happened?

What the LLM sees

Next Steps

On this page