Ollama-and-LangChain-MCP-adapters

Ollama and LangChain MCP Adapters

A local Model Context Protocol (MCP) implementation that demonstrates how to build a secure, private MCP client using Ollama for local LLM hosting and LangChain for AI framework integration. This project showcases database management through MCP tools with a SQLite backend.

Introduction

This project implements a complete MCP (Model Context Protocol) ecosystem that enables AI models to interact with databases through standardized tools. Unlike cloud-based solutions like Cursor IDE or Claude Desktop, this implementation runs entirely locally, making it perfect for handling sensitive data that requires privacy and security.

The system consists of:

• MCP Server: Exposes database operations as standardized tools
• MCP Client: Connects to the server using LangChain and Ollama
• Local LLM: Uses Ollama to run models locally (llama3.2)
• Database: SQLite database for storing and retrieving people records

How It Works

Architecture Overview

User Input → LangChain Agent → MCP Tools → SQLite Database
     ↑                                           ↓
Local LLM ← Formatted Response ← Query Results ←

Key Components

1. MCP Server (server.py)

   • Implements FastMCP server with SQLite integration
   • Exposes two main tools:
     • add_data: Adds new records to the people table
     • read_data: Retrieves records from the database
   • Handles SQL query execution and error management

2. MCP Client (langchain_client.py)

   • Creates a LangChain agent with access to MCP tools
   • Uses local Ollama LLM for natural language processing
   • Provides interactive chat interface for database operations
   • Implements ReAct (Reasoning and Acting) pattern for tool usage

3. Database Schema

   CREATE TABLE people (
       id INTEGER PRIMARY KEY AUTOINCREMENT,
       name TEXT NOT NULL,
       age INTEGER NOT NULL,
       profession TEXT NOT NULL
   );
   

Workflow Example

When you ask: "Add John Doe, 30 years old, Engineer"

1. User Input → LangChain Agent processes the request
2. Agent Reasoning → Determines it needs to use add_data tool
3. SQL Generation → Creates: INSERT INTO people (name, age, profession) VALUES ('John Doe', 30, 'Engineer')
4. MCP Tool Execution → Server executes the SQL query
5. Response → Returns success confirmation to user

Getting Started

Prerequisites

• Python 3.8 or higher
• Ollama installed and running
• Git (for cloning the repository)

Installation Steps

1. Clone the repository

   git clone <repository-url>
   cd Ollama-and-LangChain-MCP-adapters
   

2. Install Ollama (if not already installed)

   # Visit https://ollama.com for installation instructions
   # For Linux/macOS:
   curl -fsSL https://ollama.com/install.sh | sh
   

3. Pull the required LLM model

   ollama pull llama3.2
   

4. Create and activate a virtual environment

   python -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate
   

5. Install Python dependencies

   pip install -r requirements.txt
   

Running the Application

1. Start the MCP Server (in one terminal)

   python server.py --server_type=sse
   

   You should see: 🚀Starting server...

2. Start the MCP Client (in another terminal)

   python langchain_client.py
   

3. Interact with the system

   Chat session started. Type 'exit' to quit.
   
   You: add Alice Smith, 28 years old, Developer
   Assistant: Successfully added Alice Smith (age: 28, profession: Developer) to the database
   
   You: show all records
   Assistant: [Displays formatted table with all records]
   
   You: exit
   

Example Commands

• Adding records:

  • "add John Doe, 30 years old, Engineer"
  • "insert Bob Wilson, 45, Manager"
  • "add new person: Sarah Johnson, age 27, profession Data Scientist"

• Reading records:

  • "show all records"
  • "list all people"
  • "display everyone in the database"
  • "show people older than 30"

Troubleshooting

Connection Issues:

• Ensure Ollama is running: ollama serve
• Check if the MCP server is accessible: curl http://127.0.0.1:8000
• Verify the correct Python environment is activated

Missing Dependencies:

pip install langchain langchain-core langchain-community langchain-mcp-adapters fastmcp langchain-ollama httpx nest-asyncio

Project Structure

Ollama-and-LangChain-MCP-adapters/
├── server.py                 # MCP server implementation
├── langchain_client.py       # MCP client with LangChain integration
├── requirements.txt          # Python dependencies
├── demo.db                   # SQLite database (created automatically)
├── communication_protocol.py # Server configuration
├── DataFlow.MD              # Detailed workflow documentation
├── article_reference.md     # Background information and references
└── README.md                # This file

Reference

For detailed background information, architectural insights, and the theoretical foundation of this implementation, please refer to article_reference.md. This file contains:

• In-depth explanation of MCP (Model Context Protocol)
• Advantages and use cases for local MCP implementations
• Detailed architecture breakdown
• Technology stack rationale
• Real-world examples and workflows

The article provides comprehensive context for understanding why and how this local MCP solution was developed, especially for scenarios requiring data privacy and security.

License

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