AI导哈

模型上下文协议服务器

一个用于将自定义工具与Claude和ChatGPT等大语言模型集成的轻量级模型上下文协议服务器实现。

  • by freshveejay
  • TypeScript
  • 本地部署
  • 更新于
在 GitHub 查看 freshveejay/mcp-server-demo

使用说明

MCP Server Demo

A minimal Model Context Protocol (MCP) server implementation for integrating custom tools with LLMs like Claude and ChatGPT.

Features

  • 🔧 Tool registration and execution
  • 📡 SSE transport for real-time communication
  • ☁️ Cloudflare Worker deployment ready
  • 🔗 Compatible with n8n, LangChain, and OpenAI function calling

Quick Start

npm install
npm run dev

Architecture

┌─────────────┐     ┌─────────────┐     ┌─────────────┐
│   Client    │────▶│  MCP Server │────▶│   Tools     │
│  (Claude)   │◀────│  (Worker)   │◀────│  (APIs)     │
└─────────────┘     └─────────────┘     └─────────────┘

Available Tools

Tool Description
html_to_pdf Convert HTML to PDF
take_screenshot Capture webpage screenshot
execute_js Run JavaScript safely

Endpoints

  • GET / - Server info
  • GET /sse - SSE connection for MCP
  • POST /tools/list - List available tools
  • POST /tools/call - Execute a tool

Deployment

Cloudflare Workers

wrangler publish

Docker

docker build -t mcp-server .
docker run -p 3000:3000 mcp-server

Integration Examples

n8n + OpenAI

  1. Register MCP endpoint as OpenAI tool
  2. Configure tool schema from /tools/list
  3. Route tool calls to /tools/call

LangChain

from langchain.tools import MCPTool

tool = MCPTool(endpoint="https://your-mcp-server.workers.dev")

License

MIT

相关 MCP