MCP Server and Client Implementation on Linux

Introduction

• Purpose of MCP: MCP aids the creation of agents and workflows using Large Language Models (LLMs).
• Integration: MCP offers pre-configured integrations for LLMs to connect to data and tools.
• Implementation Environment: Server and client implementation using Visual Studio Code with DevContainer on Linux (Debian).
• GitHub Repository: MCP-Server-and-client-Implementation-on-Linux

Project Structure

• Two directories with separate Python environments for server and client.
  • mcp_server_explore
  • mcp_client_explore

Setup Instructions

MCP Server Setup

1. Create Directory:
   • uv init mcp_server_explore
   • Navigate: cd mcp_server_explore
2. Virtual Environment:
   • Create and activate: uv venv
   • source .venv/bin/activate
3. Install Dependencies:
   • uv add "mcp[cli]"
4. Create Server File:
   • touch mcp_server_explore.py

MCP Client Setup

1. Create Directory:
   • uv init mcp_client_explore
   • Navigate: cd mcp_client_explore
2. Virtual Environment:
   • Create and activate: uv venv
   • source .venv/bin/activate
3. Install Dependencies:
   • uv add mcp python-dotenv anthropic
4. Create Client File:
   • touch mcp_client_explore.py

MCP Server Implementation

• Tools Provided:
  • add_numbers: Adds two integers.
  • multiply_numbers: Multiplies two integers.
• Server Initialization:
  • FastMCP server is initialized with FastMCP("Explore the MCP").
  • Server execution: mcp.run(transport='stdio')

MCP Client Implementation

• Client Initialization:
  • Utilizes ClientSession, StdioServerParameters, and Anthropic.
• Server Connection:
  • Connects to MCP server script (.py or .js files).
  • Lists available server tools.
• Query Processing:
  • Sends queries to Claude (LLM) and evaluates which tools to utilize.
  • Executes tool calls and manages responses.
• Interactive Chat Loop:
  • Allows user queries and tool responses until 'quit'.

Behind the Scenes

• Query Submission:
  • Tools (addition and multiplication) are listed by the client.
  • Queries are sent to Claude with tool descriptions.
  • Claude selects tools to use.
  • Tools are executed through the server, results sent to Claude.
  • Claude returns a natural language response.

Running the Application

• Commands:
  • Navigate: cd mcp_client_explore
  • Activate environment: source .venv/bin/activate
  • Run: uv run mcp_client_explore.py /workspaces/vsc_dev_mcp_explore/mcp_server_explore/mcp_server_explore.py