Introduction

What is MCP?

The Model Context Protocol (MCP) is an open standard that lets AI applications discover and call functions on your server — the same way HTTP lets browsers talk to web APIs.

Think of it as a universal plugin system for AI:

An MCP server exposes two kinds of endpoints:

• Tools — functions an LLM can execute (send an email, query a database, fetch an API). Return structured data.
• Resources — data an LLM can read (configuration, documentation, live metrics). Identified by URI, like docs://readme.

Without MCP, you'd need custom integrations for every AI app. With MCP, you write your server once and any compatible client can use it.

How It Works

┌─────────────────┐      JSON-RPC       ┌──────────────────┐      ┌─────────────┐
│   AI Agent      │ ◄──── stdio/SSE ───► │  MCP Client      │ ──►  │  Your MCP   │
│  (Claude,       │                      │  (Claude Desktop, │      │  Server     │
│   Cursor, ...)  │                      │   VS Code, ...)   │      │             │
└─────────────────┘                      └──────────────────┘      └──────┬──────┘
                                                                          │
                                                            ┌─────────────┼─────────────┐
                                                            ▼             ▼             ▼
                                                     ┌─────────┐  ┌──────────┐  ┌──────────┐
                                                     │  Tools  │  │Resources │  │  CLI     │
                                                     │ (exec)  │  │ (read)   │  │(scaffold)│
                                                     └─────────┘  └──────────┘  └──────────┘

When a user asks "What's the weather in Tokyo?", the flow is:

1. The AI agent decides it needs weather data
2. It asks the MCP client for available tools
3. The client sends a tools/list request to your server over stdio
4. Your server responds with the get_weather tool definition (name, description, input schema)
5. The AI agent sees the tool and decides to call it with { city: "Tokyo" }
6. The client sends a tools/call request to your server
7. Your run handler executes, fetches live weather data, and returns it
8. The AI agent receives the response and crafts a natural language answer

Why mcpcraft-sdk?

The official @modelcontextprotocol/sdk gives you full control — but 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()

CLI, SDK & Templates

mcpcraft-sdk ships as three integrated layers:

The CLI is optional — you can use the SDK directly by installing the package. Templates are for when you want to skip the setup and start coding immediately.

Features

Zero Boilerplate

Define tools with plain objects. No JSON-RPC handlers, no transport setup, no manual schema definitions.

Full Type Safety

Handler parameters are typed automatically from your input schema — full autocomplete, zero casts.

Built-in Validation

Every tool input is validated with Zod before reaching your handler. Invalid calls get structured error responses, not crashes.

Tools & Resources

Register executable tools (for LLM actions) and read-only resources (for context) with the same .add() API.

Production Ready

Built on top of the official @modelcontextprotocol/sdk. Fully spec-compliant, works with Claude Desktop, VS Code Copilot, Cursor, and all MCP hosts.

Deploy Anywhere

Works with stdio (local), SSE (server-sent events), and can be deployed as Vercel serverless functions, Docker containers, or Node processes.

Next Steps

• Installation — Add mcpcraft-sdk to your project
• Quick Start — Build your first server in 5 minutes
• Deployment — Deploy to Vercel, Docker, or production
• API Reference — Full API documentation
• Examples — Real-world use cases