LangGraph ReAct Agent with MCP

This template showcases a ReAct agent implemented using LangGraph and the Model Context Protocol (MCP). The agent uses MCP servers to provide tools and capabilities through a unified gateway.

Architecture

The system consists of three main components:

1. MCP Gateway Server: A server that:

   • Manages multiple MCP server processes
   • Provides a unified API for accessing tools
   • Handles communication with MCP servers
   • Exposes tools through a simple HTTP interface

2. MCP Servers: Individual servers that provide specific capabilities:

   • Filesystem Server: File operations (read, write, list, search)
   • Memory Server: Knowledge graph operations (entities, relations, queries)
   • Additional servers can be added for more capabilities

3. ReAct Agent: The agent that:

   • Connects to the MCP gateway
   • Discovers available tools
   • Uses tools to accomplish tasks
   • Combines capabilities from multiple servers

Getting Started

1. Install Dependencies

# Install the agent package
pip install -e .

# Install the gateway package
cd gateway
pip install -e .
cd ..

2. Configure MCP Servers

The gateway server is configured through gateway/config.json. By default, it starts two MCP servers:

{
  "mcp": {
    "servers": {
      "filesystem": {
        "command": "npx",
        "args": [
          "-y",
          "@modelcontextprotocol/server-filesystem",
          "/path/to/directory"
        ]
      },
      "memory": {
        "command": "npx",
        "args": [
          "-y",
          "@modelcontextprotocol/server-memory"
        ]
      }
    }
  }
}

You can add more servers from the official MCP servers repository.

3. Start the Gateway Server

cd gateway
python -m mcp_gateway.server

The server will start on port 8808 by default.

4. Configure the Agent

The agent's connection to the gateway is configured in langgraph.json:

{
  "dependencies": ["."],
  "graphs": {
    "agent": "./src/react_agent/graph.py:graph"
  },
  "env": ".env",
  "mcp": {
    "gateway_url": "http://localhost:8808"
  }
}

5. Use the Agent

Open your app in LangGraph! Install guide here.

This will open a new browser window with the agent running. The agent will automatically:

1. Connect to the local gateway server
2. Discover available tools
3. Make tools available for use in conversations

Available Tools

The agent has access to tools from both MCP servers:

Filesystem Tools

• read_file: Read file contents
• write_file: Create or update files
• list_directory: List directory contents
• search_files: Find files matching patterns
• And more...

Memory Tools

• create_entities: Add entities to knowledge graph
• create_relations: Link entities together
• search_nodes: Query the knowledge graph
• And more...

Development

Adding New MCP Servers

1. Find a server in the MCP servers repository
2. Add its configuration to gateway/config.json
3. The agent will automatically discover its tools

Customizing the Agent

• Modify the system prompt in src/react_agent/prompts.py
• Update the agent's reasoning in src/react_agent/graph.py
• Add new capabilities by including more MCP servers

Documentation

• LangGraph Documentation
• Model Context Protocol
• MCP Servers Repository

License

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