Implementing a client

Overview

• The MCP client wraps a client session, which is the actual connection to the MCP server
• The client session is part of the MCP Python SDK
• The client class exists primarily to manage resource cleanup (connection lifecycle)
• Cleanup code is handled in the connect function and async enter/async exit functions

Purpose of the Client

• Exposes server functionality to the rest of the codebase
• Allows other code to access tools and prompts defined by the MCP server
• Acts as an intermediary between the application and the MCP server

Key Functions to Implement

list_tools()

• Calls self. session. list_tools() to retrieve all available tools from the server
• Returns result. tools containing tool definitions
• Each tool includes a description and input schema
• These definitions are passed to Claude

call_tool()

• Calls self. session. call_tool(tool_name, tool_input)
• Takes the tool name and input parameters provided by Claude
• Executes the tool on the server and returns the result

Testing the Implementation

• Use the testing harness at the bottom of the mcp_client. py file
• Run with uv run mcp_client. py (or python mcp_client. py without UV)
• Verify that tools are listed correctly and can be called
• Once implemented, the CLI can request Claude to use these tools to read and edit documents

Takeaways

• MCP client wraps a client session to manage the connection to the MCP server and handle resource cleanup
• The client exposes server functionality by allowing other code to access tools and prompts defined by the MCP server
• list_tools() retrieves available tools from the server via self. session. list_tools() and returns tool definitions with descriptions and input schemas
• call_tool() executes a tool on the server by calling self. session. call_tool(tool_name, tool_input) with parameters provided by Claude
• Test the implementation using the testing harness with uv run mcp_client. py to verify tools are listed and callable