📚Academy
likeone
online
Module 1 · Lesson 2

MCP Architecture

Three components, one protocol. Understand how Hosts, Clients, and Servers form the MCP communication layer.

The Three Components

Click each component to explore what it does in the MCP architecture:

🧠

Host

Claude, ChatGPT, or any LLM application

🌐

MCP Client

The bridge between Host and Server

⚙️

MCP Server

Your tool or data source

Host Application

The Host is the AI application the user interacts with — like Claude Desktop, Claude Code, or a custom app built with the Anthropic SDK. The Host initiates MCP connections and decides which servers to connect to. It contains one or more MCP Clients.

Role: Receives user input, decides when tools are needed, orchestrates the overall interaction.

Claude Desktop Claude Code Custom SDK apps IDE extensions

Architecture Diagram

Watch how data flows through the MCP architecture:

Data Flow — Step by Step

Click play to animate the request lifecycle, or click individual steps:

1
User asks a question"What files are in my project directory?"
2
Host realizes it needs external dataThe LLM sees available tools and decides to use the filesystem tool
3
Client sends request to ServerMCP Client sends a tools/call request via JSON-RPC to the filesystem server
4
Server executes and returns dataFilesystem server reads the directory and returns the file listing
5
Host generates an informed answerThe LLM uses the real file data to respond accurately to the user

Put the Data Flow in Order

Arrange these steps in the correct order for an MCP request

1Host identifies a tool is needed
2Host generates an informed answer
3Server executes and returns data
4User asks a question
5MCP Client sends tools/call to Server

Match Component to Role

Tap one on the left, then its match on the right

MCP Architecture — Console
Free response

Describe the MCP architecture in words: Client, Server, and Tools. Explain what each component does and how they communicate.

Type a prompt below to get started.

Try:

Academy
Built with soul — likeone.ai