MCP Blog Project

This project demonstrates the implementation of Model Context Protocol (MCP) in a blog application system. It consists of three main components that work together to provide a seamless integration between language models and structured data operations.

Project Structure

.
├── books-mcp-server/     # TypeScript MCP server implementation
│   ├── build/           # Compiled JavaScript files
│   ├── src/             # Source TypeScript files
│   └── package.json     # Server dependencies and scripts
├── books-mock-service/  # Mock service for book data
│   └── server.js        # Express.js mock server
└── mcp-client/         # Python MCP client implementation
    └── client.py       # Main client application

Components

1. Books MCP Server

A TypeScript-based MCP server that handles the communication protocol between the client and the mock service.

• Built with TypeScript for type safety
• Uses `@modelcontextprotocol/sdk` for MCP implementation
• Runs on stdio transport layer

2. Books Mock Service

A simple Express.js server that simulates a book data service.

• Provides mock endpoints for book-related operations
• Built with Express.js
• Lightweight and easy to extend

3. MCP Client

A Python-based client that integrates with language models using the MCP protocol.

• Uses `langchain_ollama` for LLM integration
• Implements MCPAgent for handling queries
• Provides an interactive CLI interface

Setup Instructions

1. Books MCP Server Setup

cd books-mcp-server
npm install
npm run build

2. Books Mock Service Setup

cd books-mock-service
npm install
npm run dev

3. MCP Client Setup

cd mcp-client
# Create and activate virtual environment (recommended)
python -m venv .venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate
# Install dependencies (requirements.txt needed)
pip install langchain-ollama mcp-use

Usage

1. Start the mock service:

cd books-mock-service
npm run dev

2. In a new terminal, run the client:

cd mcp-client
python client.py

3. Enter your queries when prompted. Type 'exit' to quit.

Development

• The MCP server is built using TypeScript. Make changes in the `src` directory and rebuild using `npm run build`
• The mock service can be extended by adding new routes in `server.js`
• The client can be customized by modifying the LLM configuration in `client.py`

Dependencies

• Node.js and npm for the MCP server and mock service
• Python 3.x for the client
• Ollama for local LLM support
• Various npm packages as specified in package.json files
• Python packages: langchain-ollama, mcp-use