AI-Tool-Calling-Agent-with-LangChain-MCP

# AI-Tool-Calling-Agent-with-LangChain-MCP # 🧠 AI Tool-Calling Agent with LangChain & MCP An autonomous AI agent built using **LangChain’s ReAct framework** and **Model Context Protocol (MCP)**. This project demonstrates how an LLM can reason and interact with external tools like math operations and weather forecasting through tool-calling. --- ## 🚀 Project Overview This project showcases how to build an **AI agent that can "think" and "act"** by: - **Interpreting user questions** - **Deciding what tool to use** - **Calling external tools** (like a calculator or weather app) - **Returning a final answer**, combining LLM reasoning with tool outputs --- ## 🛠️ Features ✅ Tool-calling agent powered by **LangChain ReAct Agent** ✅ Custom tools defined and served via **MCP Server** ✅ Uses **asynchronous MCP client** to fetch tool capabilities ✅ Seamlessly integrates tools like `add`, `multiply`, and `get_weather_forecast(city)` ✅ Runs on **Groq LLaMA 3 70B Versatile** via `langchain-groq` --- ## 📦 Tech Stack | Layer | Tool / Library | |------------------|----------------------------------------------| | 🧠 LLM | Groq - LLaMA 3 70B Versatile | | 🧩 Agent | LangChain - ReAct Agent | | 🔗 Tool Protocol | MCP (Model Context Protocol) | | 🔌 Tools Adapter | `langchain_mcp_adapters` | | 🔁 Async Client | `ClientSession` + `stdio_client` (MCP) | | 🐍 Language | Python 3.10+ | --- ## 🧪 Example Prompt What's the current weather condition in Chennai? Behind the scenes: The LLM receives the prompt. It decides to call the get_weather_forecast(city) tool. Tool gets invoked from the MCP Server. Result is returned and combined into a natural response by the agent. How It Works 🔨 Step 1: Define MCP Tools # math_server.py @math_mcp.tool() def add(a: int, b: int) -> int: return a + b @math_mcp.tool() def multiply(a: int, b: int) -> int: return a * b 🔗 Step 2: Connect AI Agent to MCP Server async with stdio_client(server_parameters) as (read, write): async with ClientSession(read, write) as session: await session.initialize() mcp_tools = await load_mcp_tools(session) ai_agent = create_react_agent(llm_model, mcp_tools) response = await ai_agent.ainvoke({"messages": prompt}) 🧭 Running the Project Start MCP Server: python mcp_server.py Run the AI Agent: python run_agent.py 🧰 Tool Debugging with MCP Inspector (Optional) Install and run the visual MCP Inspector: npx mcp inspector This launches a local web app where one can test the tool interface visually. ✔ Learnings and Key takeaways from Implementing this Simple Agent: - Understanding of LangChain's ReAct Agent architecture - Tool-Calling design using Model Context Protocol (MCP) - Practical use of async programming and agent-to-tool messaging - Combining LLMs + Tools to create semi-autonomous AI systems 📌 Credits LangChain Model Context Protocol Groq - blazing fast inference LangChain MCP Adapters