🧩 MCP Demo with LangChain, LangGraph & UV

This project demonstrates how to build custom MCP (Model Context Protocol) servers and integrate them with a LangGraph ReAct agent using LangChain MCP adapters. All dependencies are managed using the UV package manager.

🖼️ MCP Host–Server Architecture

What this shows:

• An MCP Host (Claude, Cursor, or your own app) runs MCP clients.
• Each MCP client maintains a 1:1 JSON-RPC connection with an MCP server.
• MCP servers act as adapters to:
  • Private APIs
  • Databases
  • Public web APIs

In this project, the LangGraph agent acts as the MCP host.

🖼️ MCP Servers, Clients & Agent Interaction

Explanation:

• Each MCP server can expose multiple tools.
• MCP clients live inside the host application.
• The LangGraph ReAct agent dynamically loads MCP tools and decides when to invoke them.
• Your Python MCP client uses load_mcp_tools internally via adapters.

This maps directly to:

• mathserver.py → Math tools
• weather.py → Weather tool

🖼️ Model Context Protocol (MCP) Flow

Flow explained step-by-step:

1. User asks a question
2. LLM reasons inside the agent
3. MCP Client sends a standardized request
4. MCP Server fetches or executes logic
5. Data is returned back through MCP
6. Agent responds with final output

📁 Project Structure

mcp-demo-langchain/
│
├── client.py            # MCP client + LangGraph ReAct agent
├── mathserver.py        # Math MCP server (stdio)
├── weather.py           # Weather MCP server (streamable-http)
│
├── requirements.txt
├── pyproject.toml
├── uv.lock
├── README.md
├── .gitignore
├── .env
└── .venv/

⚙️ Setup Using UV

pip install uv
uv venv
source .venv/bin/activate
uv pip install -r requirements.txt

🚀 Running the Project

Start Weather MCP Server

python weather.py

Run MCP Client & Agent

python client.py

🧪 Example Outputs

Math

User: what's (3 + 5) x 12?
Output: 96

Weather

User: what is the weather in California?
Output: It's always raining in California

✨ Key Takeaways

• MCP standardizes tool communication for LLMs
• MCP servers are reusable, language-agnostic adapters
• LangGraph enables structured agent reasoning
• UV provides fast, reproducible dependency management

📄 License

For educational and demonstration purposes.