MCP 服务器

Dynamic LangGraph MCP Agent

A production-ready agent system that automatically discovers and uses tools from MCP servers using LangGraph's ReAct architecture and LLM-powered routing.

README

🚀 Dynamic LangGraph MCP Agent

A production-ready agent system that automatically discovers and uses tools from MCP (Model Context Protocol) servers using LangGraph's ReAct architecture.

✨ Features

🤖 LangGraph ReAct Agent - Built-in reasoning and multi-step planning
🔍 Automatic Tool Discovery - No hardcoding, just add MCP servers and go
🧠 LLM-Powered Routing - Gemini Flash intelligently selects the right tools
🔌 Multi-Server Support - Connect to unlimited MCP servers
📝 Multi-Step Reasoning - Agent can chain multiple tools to solve complex tasks
✨ Zero Configuration - Add tools and they work instantly

📁 Project Structure

mcp-agent/
├── agents.py              # Main application (FastAPI + LangGraph)
├── mcp_server.py          # MCP server with agricultural tools
├── config.json            # MCP server configuration
├── .env                   # Environment variables (API keys)
├── requirements.txt       # Python dependencies
├── README.md              # This file
├── ARCHITECTURE.md        # System architecture documentation
└── DATAFLOW.md           # Complete data flow explanation

🚀 Quick Start

1. Install Dependencies

pip install -r requirements.txt

2. Set Up Environment Variables

Create a .env file:

GOOGLE_API_KEY=your_google_api_key_here

Get your API key from: https://aistudio.google.com/app/apikey

3. Configure MCP Servers

Edit config.json with your MCP server paths:

{
  "mcpServers": {
    "agricultural-server": {
      "command": "python",
      "args": ["mcp_server.py"],
      "env": {
        "PYTHONIOENCODING": "utf-8"
      }
    }
  }
}

Important: Use full paths on Windows:

{
  "command": "D:\\Python\\python.exe",
  "args": ["D:\\projects\\mcp-agent\\mcp_server.py"]
}

4. Start the Server

python agents.py

5. Test the Agent

Visit http://localhost:8000/docs

Or use cURL:

curl -X POST "http://localhost:8000/chat" \
  -H "Content-Type: application/json" \
  -d '{"message": "What is the weather in Tokyo?"}'

📊 Available Tools

Tool	Description	Arguments
`get_current_weather`	Real-time weather data	city (string)
`get_pesticide_seed_info`	Agricultural information	query (string)
`get_placeholder_posts`	Sample blog posts	limit (integer)

🧪 Example Queries

# Weather query → Uses get_current_weather
"What's the weather in Paris?"

# Agriculture query → Uses get_pesticide_seed_info  
"Tell me about organic farming techniques"

# Content query → Uses get_placeholder_posts
"Show me 5 interesting articles"

# Multi-step reasoning → Uses multiple tools
"What's the weather in Mumbai and what crops grow best there?"

🔧 API Endpoints

`POST /chat`

Main endpoint for chatting with the agent

Request:

{
  "message": "Your query here"
}

Response:

{
  "response": "Agent's answer",
  "intermediate_steps": ["Tool used: get_current_weather"],
  "error": null
}

`GET /` - Server info

`GET /tools` - List all tools

`GET /health` - Health check

🔌 Adding New Tools

Edit mcp_server.py:

@mcp_server.list_tools()
async def list_tools() -> list[Tool]:
    return [
        # ... existing tools ...
        Tool(
            name="my_new_tool",
            description="What this tool does",
            inputSchema={
                "type": "object",
                "properties": {
                    "param": {"type": "string"}
                },
                "required": ["param"]
            }
        )
    ]