Implementing a client

MCP Client Implementation Notes

**I. MCP Client Overview**

• Purpose: The MCP client (located in mcp-client. py) acts as a wrapper around a client session, providing an interface to interact with the MCP server.
• Dual Role: This project simultaneously implements both a client and a server to demonstrate both sides of the communication puzzle.
• Client Session: This is the actual connection to the MCP server, provided by the MCP Python SDK.
• Client Class Function: The main client class manages the client session and handles necessary resource cleanup (e. g. , using async enter and async exit functions).
• Client Role: The client exposes server functionality (like tools and prompts) to the rest of the codebase.

**II. Core Client Functionality**

• Client Functions: The client provides methods to interact with the server, such as list_tools, call_tool, list_prompts, and get_prompt.
• list_tools Implementation:
• Calls self. session. list_tools().
• Returns the list of tools defined by the server.
• call_tool Implementation:
• Calls self. session. call_tool(tool_name, tool_input).
• Passes the tool name and input arguments (provided by the calling code, e. g. , Claude) to the server.

**III. Testing and Integration**

• Testing Harness: A testing block at the bottom of mcp-client. py allows direct execution of the client to verify connectivity and functionality.
• Tool Definition Gotcha:
• The MCP specification's tool definition structure is slightly different from the structure expected by Bedrock JSON schema.
• The code must include a translation step (e. g. , the to_bedrock_tools function in bedrock. py) to convert the MCP tool definition into the required Bedrock format.
• End-to-End Workflow:
• The client lists tools from the server.
• The tools are translated for Bedrock.
• The calling code (e. g. , the CLI) sends the tools to an LLM (Claude).
• Claude decides to use a tool (e. g. , read_document).
• The client executes the tool call on the server, retrieving the required data.

Takeaways

• The MCP client acts as a wrapper around the MCP Python SDK session, providing an interface to interact with the MCP server and managing resource cleanup.
• Core client functions, such as list_tools and call_tool, are used to facilitate communication between the client and the server.
• A necessary translation step (e. g. , to_bedrock_tools) must convert the MCP tool definition structure into the required Bedrock JSON schema format.
• The end-to-end workflow involves the client listing tools, translating them for the LLM, and then executing the tool call back on the server.