AI导哈

AI代理开发工具包

一个基于模型上下文协议（MCP）和OpenAI的完整工具包，用于构建自定义AI代理，实现现实世界工具与大型语言模型（LLM）的连接。

  • by thusharanavod123
  • JavaScript
  • 本地部署
  • 2 星标
  • 更新于
在 GitHub 查看 thusharanavod123/mcp-agent-starter-kit

使用说明

🧠 MCP代理入门套件

🚀 使用以下工具构建自定义AI代理的完整入门套件 模型上下文协议（MCP） 和 开放人工智能.
学习如何创建、运行和扩展智能代理，这些代理可以将现实世界的工具与LLM连接起来。

🧩 这是什么？

这个项目是 最小但功能强大的模板 这有助于开发人员使用 模型上下文协议（MCP） --实现工具和大型语言模型（LLM）之间结构化交互的新标准。

它附带了：

  • ✅ 基本的 MCP服务器
  • ✅ 一个样本 AI工具(WeatherTool)
  • ✅ 与集成 OpenAI API
  • ✅ A工作 客户端示例 演示工具调用

🏗️ 项目结构

mcp试剂入门套件/ │ ├── index.js#主客户端入口点（OpenAI+MCP） ├── mcp/ │ ├── server.js#MCP服务器设置 │ └── 工具/ │ └── weatherTool.js#自定义工具示例 │ ├── .env#OpenAI API密钥 ├── package.json └── README.md

⚙️ 安装说明

1.️⃣ 克隆存储库

git clone https://github.com/thusharanavod123/mcp-agent-starter-kit.git
cd mcp-agent-starter-kit

2️⃣ Install Dependencies
npm install

3️⃣ Configure Environment Variables

Create a .env file in the root directory:

OPENAI_API_KEY=your_openai_api_key_here

4️⃣ Run the MCP Server
node mcp/server.js


✅ You should see:

MCP-like server running on http://localhost:8080

5️⃣ Test Your MCP Agent

If you’re using PowerShell (Windows):

Invoke-WebRequest -Uri "http://localhost:8080/run-tool" -Method POST -Body '{"city":"Colombo"}' -ContentType "application/json"


If you’re using Mac/Linux terminal:

curl -X POST http://localhost:8080/run-tool \
  -H "Content-Type: application/json" \
  -d '{"city":"Colombo"}'


✅ You’ll get a JSON response like:

{"result": "Weather in Colombo is 28°C and sunny."}

🧠 How It Works

MCP Server (mcp/server.js)

Hosts your custom AI tools (e.g., WeatherTool).

Listens on a local port (default: 8080).

MCP Tools (mcp/tools/)

Define real-world functions (e.g., fetching weather, sending emails, running commands).

Each tool can expose name, description, and execute().

OpenAI Client (index.js)

Connects your LLM with your MCP server.

Sends user queries → triggers MCP tools → returns intelligent responses.

🧩 Example Custom Tool

Example: mcp/tools/weatherTool.js

export const WeatherTool = {
  name: "WeatherTool",
  description: "Fetches current weather for a given city",
  async execute({ city }) {
    return { result: `Weather in ${city} is 28°C and sunny.` };
  },
};
You can replace this with any tool (e.g., file searcher, calendar integration, API fetcher).

🧭 Branch Structure
Branch	Purpose
main	Stable version ready for production
dev	Experimental branch for testing new tools and features
agent-core	MCP core logic improvements
tool-plugins	For building and testing modular tools (e.g., calculator, email sender)
docs	Documentation, examples, and tutorials
🌍 Topics
mcp ai-agent model-context-protocol openai nodejs ai-framework ai-tools llm mcp-server developer-tools

🤝 Contributing

Contributions are welcome!
Fork the repo, make your changes, and submit a PR.
If you build an awesome custom tool, feel free to add it under /mcp/tools/ and share it with the community.

📜 License

This project is licensed under the MIT License
.

⭐ Support

If this repo helps you, please star ⭐ it and share it on:

LinkedIn

X (Twitter)

Dev.to

Every star helps inspire new developers to explore MCP and AI tools. 🌱



---

Would you like me to also create a **`CONTRIBUTING.md`** and **GitHub banner idea (header image prompt)** so your repo looks visually stunning and professional on GitHub?

相关 MCP