Since Anthropic introduced the Model Context Protocol (MCP) in November 2024, many resources have explored its concepts and applications.This article aims to offer a lower-level, multi-server example without using FastMCP to delve into MCP mechanisms.MCP facilitates interactions between large language models (LLMs) and diverse tools beyond training data.The MCP ecosystem involves servers hosting tools, clients connecting to servers, and hosts coordinating interactions.The article demonstrates implementing this architecture using lower-level Python MCP SDK classes, bypassing high-level wrappers.Servers are defined with tools like MathServer, and connections are established using SSE transport protocol.Additional libraries like starlette and uvicorn are used in the examples for building multi-server setups.A MultiServerClient class manages connections to multiple servers, handling communication and tool routing.The host component interacts with a language model to decide which tools to invoke based on user queries.The article provides a clear example of how to implement MCP with multiple servers in a structured manner, inviting further exploration.