MCP Client

A Python client for connecting to Machine-to-Machine Communication Protocol (MCP) servers. This client allows you to interact with both stdio and SSE MCP servers, enabling LLM agents to use tools through a standardized protocol.

Features

• Connect to stdio MCP servers (Python and JavaScript)
• Connect to SSE MCP servers
• Interactive chat interface with multiple LLM options:
  • Anthropic Claude 3.5 Sonnet (default)
  • OpenAI GPT-4o
  • Google Gemini 2.0 Flash
• Tool calling with automatic result processing
• Conversation history management with refresh capability
• Detailed logging for debugging and monitoring

Prerequisites

• Python 3.8+
• API keys set as environment variables:
  • ANTHROPIC_API_KEY for Anthropic Claude
  • OPENAI_API_KEY for OpenAI GPT models
  • GOOGLE_API_KEY for Google Gemini

Installation

1. Clone the repository
2. Install the required packages:

pip install mcp-protocol-client anthropic openai google-genai python-dotenv

3. Create a .env file with your API keys:

ANTHROPIC_API_KEY=your_anthropic_api_key
OPENAI_API_KEY=your_openai_api_key
GOOGLE_API_KEY=your_google_api_key

4. Create a logs directory to store client logs:

mkdir logs

Usage

Basic Usage

python client.py <server_script_path_or_url> [llm_provider]

Where:

• <server_script_path_or_url> is either a path to an MCP server script or URL to an SSE MCP server
• [llm_provider] is optional and can be one of: anthropic (default), openai, or gemini

Examples

1. Connect to a Python MCP server using Anthropic Claude (default):

python client.py ./weather.py

2. Connect to a JavaScript npm MCP server using OpenAI:

python client.py @playwright/mcp@latest openai

3. Connect to an SSE MCP server using Google Gemini:

python client.py http://localhost:8000/sse gemini

Interactive Chat Commands

• Type your queries to interact with the LLM and tools
• Type refresh to clear conversation history
• Type quit to exit the application

Development

The client includes VS Code launch configurations for various setups, making it easy to debug and test with different servers and LLM providers.

How It Works

The MCP Client:

1. Connects to an MCP server (either stdio or SSE)
2. Lists available tools from the server
3. Processes user queries by:
   • Sending the query to the selected LLM with available tools
   • Detecting and executing tool calls when the LLM requests them
   • Sending tool results back to the LLM for processing
   • Providing the final response to the user
4. Maintains conversation history for context

Logging

Logs are stored in logs/mcp_client.log and are also displayed in the console. The logging level can be adjusted in the client.py file.