MCP LangChain

A project demonstrating how to integrate Model Context Protocol (MCP) with LangChain and LangGraph to create AI agents that can use custom tools.

What This Project Does

This project shows how to:

• Create custom MCP tool servers (math operations, weather lookup)
• Connect an LLM (Ollama/Groq) to these tools using LangChain
• Use the ReAct (Reason + Act) agent pattern for intelligent tool usage

In Simple Terms

Built a system where:

1. MCP Servers expose Python functions as "tools" (like a calculator, weather API)
2. LangChain connects these tools to an LLM (AI model)
3. LangGraph's ReAct Agent helps the LLM decide when and how to use these tools
4. Ollama runs a free, local AI model.

Project Structure

mcp-lang/
├── client.py              # Main client - connects LLM to MCP servers
├── servers/
│   ├── mathserver.py      # MCP server with math tools (stdio transport)
│   └── weatherserver.py   # MCP server with weather tool (HTTP transport)
├── main.py                # Basic entry point
├── requirements.txt       # Python dependencies
├── pyproject.toml         # Project configuration
└── .env                   # Environment variables (API keys)

How It Works (ReAct Pattern)

1. User asks a question → "What is 5 multiplied by 3?"
2. LLM reasons → "This is a math problem, I should use multiply_two_numbers tool"
3. LLM acts → Calls the tool with arguments {"a": 5, "b": 3}
4. Tool executes → MCP server returns 15
5. LLM responds → "5 multiplied by 3 is 15"

---

Installation & Setup

Prerequisites

• Python 3.14+
• Install Ollama for local LLM

Option 1: Direct Download: Go to https://ollama.ai/download and Download the macOS installer

Option 2: Using Homebrew: brew install ollama

Also to install local llm models
Pull the Model: `ollama pull llama3.1:8b`

Alternative Models (that support tools):
Smaller, faster (needs less RAM): ollama pull llama3.2:3b
Larger, smarter (needs more RAM): ollama pull llama3.1:70b
Mistral (good alternative): ollama pull mistral:7b

Step 1: Clone the Repository

Step 2: Create Virtual Environment

# Using Python venv
python -m venv .venv
# Activate virtual environment
# On macOS/Linux:
source .venv/bin/activate
# On Windows (Command Prompt):
.venv\Scripts\activate
# On Windows (PowerShell):
.venv\Scripts\Activate.ps1

Step 3: Install Dependencies

pip install -r requirements.txt

uv pip install -r requirements.txt #Or using uv (faster)

Running the Project

Step 1: Start the Weather Server (HTTP Transport)

In a separate terminal:

cd mcp-lang
source .venv/bin/activate  # or .venv\Scripts\activate on Windows
python servers/weatherserver.py

Keep this terminal running. You should see:

Starting MCP server on http://localhost:8000/mcp

Step 2: Run the Client

In a new terminal:

cd mcp-lang
source .venv/bin/activate
python client.py

Expected Output

============================================================

Response The current weather in New York is sunny with a temperature of 25°C.

============================================================

Response 5 multiplied by 3 is 15.

============================================================

---

Code Explanation

client.py - The Main Application

# Import the MCP client adapter for LangChain
from langchain_mcp_adapters.client import MultiServerMCPClient

# Import the ReAct agent pattern from LangGraph
from langgraph.prebuilt import create_react_agent

# Import the Ollama LLM wrapper
from langchain_ollama import ChatOllama

Key Concepts:

1. MultiServerMCPClient: Connects to multiple MCP tool servers simultaneously
2. create_react_agent: Creates a ReAct agent that reasons about when to use tools
3. ChatOllama: Wrapper to use local Ollama models with LangChain

servers/mathserver.py - Math Tool Server

from mcp.server.fastmcp import FastMCP

mcp = FastMCP("mathserver")

@mcp.tool("add_two_numbers")
def add(a: float, b: float) -> float:
    """Add two numbers together."""
    return a + b

Key Concepts:

1. FastMCP: Quick way to create MCP servers
2. @mcp.tool(): Decorator to expose functions as tools
3. Type hints: Required for LLM to understand parameters

Transport Types

Transport	Use Case	Server Start
stdio	Client spawns server as subprocess	Automatic
streamable-http	Server runs independently on HTTP	Manual

---

This project is for learning purposes