🤖 MCP Agent Framework

A modular system for creating AI agents using the Model Context Protocol (MCP). This framework allows connecting to multiple MCP servers simultaneously, aggregating tools, and executing queries using LangChain agents.

🌟 Features

• 🔌 Multi-server Connection: Connect to multiple MCP servers simultaneously
• 🛠️ Tool Aggregation: Automatically collect and combine tools from all connected servers
• 🤖 LangChain Integration: Create React agents using the LangChain library
• 🔄 Asynchronous Processing: Fully asynchronous code with proper resource management
• 🧠 OpenAI Integration: Seamless integration with OpenAI models for intelligent agent responses
• 📊 Detailed Logging: Comprehensive logging of all operational stages for debugging
• 🧩 Modular Architecture: Clear separation into client and agent components

Demo

cursor.mp4

terminal_demo.mp4

🚀 Quick Start

1. Installation

# Clone the repository
git clone https://github.com/extrawest/agent_with_multiple_mcp_servers.git
cd agent_with_multiple_mcp_servers

# Install dependencies
pip install -r requirements.txt

2. Configuration

Copy the example environment file and configure your settings:

cp .env.example .env

3. Start MCP Servers

Before running the agent, you need to start the MCP servers. The project includes two different servers:

Stocks Server

Provides financial data tools using yfinance:

# Start the stocks server on port 7860
python -m mcp_components.server.stocks_server

News Server

Provides news retrieval tools using NewsAPI:

# Start the news server on port 7861
python -m mcp_components.server.news_server

4. Run the Agent

Once the servers are running, you can start the agent:

# Run the agent with connections to both MCP servers
python run_agent.py

5. Cursor IDE Integration

You can add the MCP servers to Cursor IDE by adding the following configuration to your Cursor settings:

{
  "mcpServers": {
    "stocks-mcp-server": {
      "url": "http://127.0.0.1:7860/sse"
    },
    "news-mcp-server": {
      "url": "http://127.0.0.1:7861/sse"
    } 
  }
}

Make sure both servers are running before connecting through Cursor IDE.

📦 Project Structure

mcp_components/
├── agent/
│   └── agent.py       # AgentRunner implementation for working with MCP servers
├── client/
│   └── client.py      # Client for connecting to MCP servers
├── server/
│   ├── stocks_server.py  # MCP server providing stock market tools
│   └── news_server.py    # MCP server providing news retrieval tools

💻 Usage

Direct MCP Client Usage

from mcp_components.client.client import MCPClientWrapper
from typing import List, Any, Tuple

async def get_mcp_client(server_url: str) -> Tuple[MCPClientWrapper, List[Any]]:
    client = MCPClientWrapper(server_url)
    await client.connect()
    tools = await client.load_tools()
    return client, tools

🔧 Components

AgentRunner

Class for creating and managing agents that can use tools from multiple MCP servers:

• setup_with_multiple_mcp_servers(): Connect to multiple MCP servers and aggregate tools
• run_query(): Execute a query using the agent
• cleanup(): Close all connections and free resources

MCPClientWrapper

Class for connecting to MCP servers and loading tools:

• connect(): Establish a connection with an MCP server
• load_tools(): Load available tools from the server
• close(): Close the connection to the server

MCP Servers

The framework includes two specialized MCP servers:

Stocks Server

• Provides financial data tools using yfinance
• Runs on port 7860 by default
• Tools:
  • fetch_stock_info: Get general company information
  • fetch_quarterly_financials: Get quarterly financial data
  • fetch_annual_financials: Get annual financial data

News Server

• Provides news retrieval tools using NewsAPI
• Runs on port 7861 by default
• Tools:
  • fetch_news: Get news articles related to a query

📋 Requirements

• Python 3.9+
• MCP 1.9.3+
• LangChain
• OpenAI API key (for using OpenAI models)
• NewsAPI key (for the news server)
• yfinance (for the stocks server)

🔍 Implementation Details

• Asynchronous Context Managers: Proper management of asynchronous resources using AsyncExitStack
• Error Handling: Robust exception handling at all levels
• Connection Preservation: Maintaining active connections for tool usage
• Tool Aggregation: Combining tools from different servers into a unified list

Developed by extrawest. Software development company