AI导哈

多代理工具调用系统

一个基于LangChain、Groq和MCP的多代理系统，支持动态工具发现和执行，适用于智能代理工具集成和任务自动化场景。

  • by Saketh-Vadlamudi
  • Python
  • 本地部署
  • 更新于
在 GitHub 查看 saketh-vadlamudi/mcp

使用说明

使用LangChain、Groq和MCP进行多代理工具调用

使用LangChain、Groq的Llama3模型和多服务器通信协议（MCP）演示动态工具发现和执行的多代理系统。该项目展示了如何将各种工具服务器（数学和天气）与不同的通信传输（stdio和HTTP流式传输）集成，以供智能代理使用。

特性

  • 多服务器通信协议（MCP）：利用MCP实现中央客户端和独立工具服务器之间的无缝通信。
  • 动态工具发现：代理客户端从注册的MCP服务器中动态发现可用工具及其功能。
  • LangChain集成：利用LangChain的 create_react_agent 用于强健的代理行为和工具编排。
  • Groq Calama3集成：使用 ChatGroq Llama3-70b-8192模型，用于强大的语言理解和响应生成。
  • 多样化的工具运输：演示两种类型的MCP传输：
    • 标准输入/输出（stdio）：对于 Math 服务器，展示简单、直接的流程通信。
    • 流式HTTP：对于 Weather 服务器，演示与 uvicorn-支持FastAPI应用程序。
  • 天气工具：使用OpenWeatherMap API获取给定位置的实时天气信息。
  • 数学工具：执行基本算术运算（加法、乘法）。
  • 错误处理：包括API密钥问题和无效城市查询的基本错误处理。

项目结构

multi-agent-mcp-project/
├── docs/                      
│   └── architecture.md
├── .gitignore                 # Files to be ignored by Git
├── client.py                  # The main agent application that orchestrates tools
├── mathserver.py              # MCP server for mathematical operations
├── requirements.txt           # Python dependencies 
├── README.md                  # Project overview, setup, usage
└── weather.py                 # MCP server for fetching weather data

安装

按照以下步骤在本地设置和运行项目：

  1. 克隆存储库：

    git clone [https://github.com/your-username/multi-agent-mcp-project.git](https://github.com/your-username/multi-agent-mcp-project.git)
cd multi-agent-mcp-project

  2. 创建虚拟环境（推荐）：

    python -m venv venv

  3. 激活虚拟环境：

    • 在macOS/Linux上：
      source venv/bin/activate
    • 在Windows上：
      .\venv\Scripts\activate

  4. 安装依赖项：

    pip install -r requirements.txt

  5. 设置环境变量：

    您需要Groq和OpenWeatherMap的API密钥。

    创建一个 .env 项目目录根目录中的文件，基于 .env.example，并用您的密钥填充它：

    cp .env.example .env

    现在，打开 .env 文件并添加您的实际API密钥：

    GROQ_API_KEY="your_groq_api_key_here"
OPENWEATHER_API_KEY="your_openweathermap_api_key_here"

    替换 "your_groq_api_key_here" 和 "your_openweathermap_api_key_here" 使用您的实际API密钥。

用法

要运行多代理系统，只需执行 client.py 脚本。这 client.py 将自动启动 mathserver.py 和 weather.py 后台的服务器并连接到它们。

python client.py
You will see output similar to this, demonstrating the agent's thought process and the results of tool calls:

🔧 Loading environment variables...
🚀 Starting main()...
🔑 GROQ_API_KEY: FOUND ✅
🌦️  OPENWEATHER_API_KEY: FOUND ✅
⚙️  Initializing MultiServerMCPClient...
🛠️  Getting tools from servers...
✅ Successfully fetched 3 tools:
   🔹 add: Add to numbers
   🔹 multiple: Multiply two numbers
   🔹 get_weather: Get the current weather for a given location. 
🧠 Initializing Groq model...
✅ Groq model initialized.
🤖 Creating React-style agent with tools...
✅ Agent created.
💬 Sending math question to agent...

--- ✅ Math Response ---
🧠 Agent Thought: The user is asking to calculate "(3 + 5) x 12". I need to use the 'add' tool first, then the 'multiple' tool.
🟢 Final Answer: 96
💬 Sending weather question to agent...

--- ✅ Weather Response ---
🧠 Agent Thought: The user is asking for the weather in "Hyderabad, India". I should use the `get_weather` tool with "Hyderabad, India" as the location.
🟢 Final Answer: 🌤️ Weather in Hyderabad, IN: overcast clouds. 29.89°C (feels like 34.02°C), 69% humidity.
💬 Sending weather query for invalid city...

--- 🧪 Weather (Invalid City) Response ---
🧠 Agent Thought: The user is asking for the weather in "NotARealCityHere?". I need to use the `get_weather` tool with "NotARealCityHere?" as the location.
🟢 Final Answer: ❌ City 'NotARealCityHere?' not found.
✅ All done!
How it Works
client.py:

Initializes MultiServerMCPClient to manage connections to different MCP servers.

Defines configurations for the Math server (using stdio transport) and the Weather server (using streamable_http transport via http://localhost:3000).

Fetches available tools from all connected servers.

Initializes the ChatGroq model.

Creates a LangChain React-style agent, providing it with the discovered tools.

Invokes the agent with various prompts to demonstrate its ability to use the add, multiple, and get_weather tools.

mathserver.py:

A simple MCP server that exposes add and multiple functions as tools.

It uses transport="stdio" to communicate via standard input/output streams.

weather.py:

An MCP server that exposes a 

get_weather function as a tool. 

It integrates with the OpenWeatherMap API to fetch weather data. 

It explicitly runs a 

uvicorn server to serve the FastMCP application via HTTP, allowing client.py to connect using streamable_http transport. 


os.environ["FASTMCP_PORT"] = "3000" is set to ensure the weather.py server starts on port 3000. 

Contributing
Contributions are welcome! Please follow these steps:

Fork the repository.

Create a new branch for your feature or bug fix.

Make your changes.

Write tests for your changes.

Ensure all tests pass.

Submit a pull request.

License
This project is licensed under the MIT License - see the LICENSE file for details.

相关 MCP