A CLI tool to create MCP (Model Context Protocol) applications with ease.
- 🚀 Quick project scaffolding
- 📦 TypeScript support out of the box
- 🛠️ Built-in development tools
- 🔧 Configurable project templates
- 🌐 Multiple Transport Modes (stdio/streamable-http/sse)
- 📚 Comprehensive APIs
npm create mcp-kit@latestor
yarn create mcp-kit@latestor
pnpm create mcp-kit@latestcreate-mcp-kit supports generating two types of projects:
Create an MCP server that provides tools, resources, and prompts for MCP clients.
The generated server project will have the following structure:
├── src/
│ ├── tools/ # MCP tools implementation
│ │ ├── index.ts # Tools registration
│ │ └── register*.ts # Individual tool implementations
│ ├── resources/ # MCP resources implementation
│ │ └── index.ts # Resources registration
│ ├── prompts/ # MCP prompts implementation
│ │ └── index.ts # Prompts registration
│ ├── services/ # Server implementations
│ │ ├── stdio.ts # STDIO transport implementation
│ │ └── web.ts # Streamable HTTP and SSE transport implementation
│ └── index.ts # Entry point
├── tests/ # Test files (optional)
├── scripts/ # Build and development scripts
├── .github/ # GitHub Actions workflows (optional)
├── .husky/ # Git hooks (optional)
└── package.json
-
npm run dev- Start the development server in stdio mode -
npm run dev:web- Start the development server in web mode -
npm run build- Build the project -
npm run test- Run tests (if vitest plugin is selected) -
npm run coverage- Generate test coverage report (if vitest plugin is selected) -
npm run lint- Run linting (if style plugin is selected)
Create an MCP client that connects to MCP servers and uses their tools, resources, and prompts.
The generated client project will have the following structure:
├── src/
│ └── index.ts # Entry point with transport implementations
├── tests/ # Test files (optional)
├── scripts/ # Build and development scripts
├── .github/ # GitHub Actions workflows (optional)
├── .husky/ # Git hooks (optional)
└── package.json
-
npm run dev- Start the client in development mode -
npm run build- Build the project -
npm run test- Run tests (if vitest plugin is selected) -
npm run coverage- Generate test coverage report (if vitest plugin is selected) -
npm run lint- Run linting (if style plugin is selected)
MCP Server supports three transport modes:
- STDIO: Communication through standard input/output streams
- Streamable HTTP: RESTful API with streaming capabilities
- SSE (Server-Sent Events): Real-time event streaming from server to client
Implement custom tools that can be used by MCP clients:
// Full implementation example
import { z } from 'zod'
import type { McpServer } from '@modelcontextprotocol/sdk/server/mcp.js'
export default function register(server, options) {
server.registerTool(
'GetData',
{
title: 'Get Data',
description: 'Get Data',
inputSchema: {
keyword: z.string().describe('search keyword'),
},
},
async ({ keyword }) => {
const { success, data, message } = await getData(keyword, options)
return {
content: [
{
type: 'text',
text: success ? data : message,
},
],
}
},
)
}
export const getData = async (keyword, options) => {
if (!keyword || keyword === 'error') {
return {
success: false,
message: 'Invalid keyword',
}
}
return {
success: true,
data: `Data for ${keyword}`,
}
}Define resources that can be accessed by MCP clients:
// Full implementation example
import { type McpServer, ResourceTemplate } from '@modelcontextprotocol/sdk/server/mcp.js'
import type { OptionsType } from '@/types'
export const registerResources = (server: McpServer, options: OptionsType) => {
server.registerResource(
'search',
new ResourceTemplate('search://{keyword}', {
list: undefined,
}),
{
title: 'Search Resource',
description: 'Dynamic generate search resource',
},
async (uri, { keyword }) => {
return {
contents: [
{
uri: uri.href,
text: `search ${keyword}`,
},
],
}
},
)
}Create reusable prompts for MCP clients:
// Full implementation example
import { z } from 'zod'
import type { McpServer } from '@modelcontextprotocol/sdk/server/mcp.js'
export const registerPrompts = (server: McpServer) => {
server.registerPrompt(
'echo',
{
title: 'Echo Prompt',
description: 'Creates a prompt to process a message.',
argsSchema: {
message: z.string(),
},
},
({ message }) => {
return {
messages: [
{
role: 'user',
content: {
type: 'text',
text: `Please process this message: ${message}`,
},
},
],
}
},
)
}Connect to MCP servers using different transport modes:
// Import the MCP client
import { McpClient } from '@modelcontextprotocol/sdk/client/mcp.js'
import { StdioClientTransport } from '@modelcontextprotocol/sdk/client/transports/stdio.js'
import { StreamableHTTPClientTransport } from '@modelcontextprotocol/sdk/client/transports/streamable-http.js'
import { SSEClientTransport } from '@modelcontextprotocol/sdk/client/transports/sse.js'
// Create a new MCP client
const client = new McpClient()
// STDIO Transport
const stdioClientTransport = new StdioClientTransport({
command: 'npx',
args: ['-y', '@my-mcp-hub/node-mcp-server'],
env: process.env,
})
await client.connect(stdioClientTransport)
// Streamable HTTP Transport
const streamableBaseUrl = new URL('http://localhost:8401/mcp')
const streamableClientTransport = new StreamableHTTPClientTransport(streamableBaseUrl)
await client.connect(streamableClientTransport)
// SSE Transport
const sseBaseUrl = new URL('http://localhost:8401/sse')
const sseClientTransport = new SSEClientTransport(sseBaseUrl)
await client.connect(sseClientTransport)Call tools provided by MCP servers:
// List available tools
const tools = await client.listTools()
console.log(tools)
// Call a tool
const callResult = await client.callTool({
name: 'GetData',
arguments: {
keyword: 'Hello',
},
})
console.log(callResult.content)Access resources provided by MCP servers:
// List available resources
const resources = await client.listResources()
console.log(resources)
// Get a resource
const resource = await client.getResource('search://example')
console.log(resource.contents)Use prompts provided by MCP servers:
// List available prompts
const prompts = await client.listPrompts()
console.log(prompts)
// Use a prompt
const prompt = await client.getPrompt('echo', { message: 'Hello, world!' })
console.log(prompt.messages)Feel free to dive in! Open an issue or submit PRs.
Standard Readme follows the Contributor Covenant Code of Conduct.
This project exists thanks to all the people who contribute.
MIT © MichaelSun
