MCP clients

Model Context Protocol (MCP) Study Notes

**I. The MCP Client**

• Purpose: Acts as the communication bridge between the main server and the MCP server.
• Function: Serves as the primary access point for all tools implemented by the MCP server.
• Transport Agnostic: The client and server can communicate over various protocols (e. g. , standard input/output, HTTP, WebSockets).

**II. Core Message Types**

• All allowed messages are strictly defined within the MCP specification.
• Tool Discovery:
• List Tools Request → Sent from Client to Server (asks for available tools).
• List Tools Result → Sent from Server to Client (provides the list of tools).
• Tool Execution:
• Call Tool Request → Sent from Client to Server (requests execution of a specific tool with arguments).
• Call Tool Result → Sent from Server to Client (contains the output/result of the tool execution).

**III. The Full Workflow (Example: Querying GitHub)**

• Initiation: A user submits a query to the main server.
• Tool Discovery Phase:
• Server requests a list of tools from the MCP Client.
• MCP Client sends a List Tools Request to the MCP Server.
• MCP Server responds with a List Tools Result.
• Execution Phase:
• Server determines a tool is needed and asks the MCP Client to run it.
• MCP Client sends a Call Tool Request to the MCP Server.
• MCP Server executes the tool (e. g. , calls GitHub API).
• MCP Server wraps the data in a Call Tool Result and sends it back to the MCP Client.
• MCP Client hands the result to the main Server.
• Final Response:
• The main Server passes the tool result (e. g. , list of repositories) to the LLM (e. g. , Claude) within a user message.
• The LLM generates a final, coherent response, which the Server sends back to the user.

Takeaways

• The MCP Client serves as the communication bridge between the main server and the MCP server, acting as the primary access point for all implemented tools.
• Communication is transport agnostic, allowing the client and server to communicate over various protocols (e. g. , HTTP, WebSockets).
• Tool discovery is initiated by the Client sending a List Tools Request to the Server, which responds with a List Tools Result.
• Tool execution involves the Client sending a Call Tool Request, and the Server returning the output via a Call Tool Result.
• The final workflow involves the main Server passing the tool result to the LLM, which then generates the coherent response for the user.