Model Context ProtocolMCP Python SDK 2.2 · LangChain 1.4 · Python 3.10+
Dashboard
0%
1
Curious builder0 XP earned · 300 to level 2
0 daysFinish a lesson to begin
Badge collection0 of 6 unlocked
27 small wins to finish your pathNext lesson →

Model Context Protocol logoMCP overview

The Model Context Protocol is an open standard that lets one AI application connect to tools and data through servers any other application can reuse.

Last updated: 29 Sep, 2026 · MCP 2.2

Anthropic introduced MCP in November 2024 and published it as an open specification with SDKs for several languages. The Python SDK used here, mcp 2.2.0, is MIT licensed and is published by the Model Context Protocol project under the Linux Foundation. This course builds one MCP server for a support team and an agent that answers with its tools.

Servers, a client and an app · from the Complete Agentic AI Course In 10 Hours · 267:35 to 271:36

The MCP section of the agentic AI course starts from one diagram with three parts. MCP servers hold tools, such as addition, multiplication or a weather API call, and a third-party company can write and run them. The MCP client keeps a one-to-one connection with a server from inside the host app. The app is Claude Desktop, or a chatbot you build yourself. Ask that app for the weather in New York and the model cannot answer from its training, so the client fetches the server's tool list, the model decides which tool to call, and the call goes to the server over the protocol.

The video builds its app with LangChain. This course builds the server and the client with the MCP Python SDK first, then brings in the video's LangChain agent once there is a server worth calling.

Hosts, clients, servers and transports

The protocol has three roles. A host is the AI application, such as Claude Code or an IDE. It creates a client for each server it connects to. A server offers three kinds of thing: tools the model can call, resources the application can read, and prompts the user can pick. Messages are JSON-RPC, carried over standard input and output for local servers or over Streamable HTTP for remote ones.

What happens when an agent uses your server
asksJSON-RPCanswersdescriptionsa tool callrunsThe modelreads descriptionsA userHost applicationa client per serverYour MCP servertools, resources, promptsYour Python functionruns on each call
Hover or tap a piece to see what it is and which lesson built it.
Trace a request

Pick one to watch it run, step by step.

Write a server once, connect it everywhere

Write an MCP server once, and Claude Code, Claude Desktop, Cursor, VS Code and your own agents can all use it, with no custom code for each one. That reuse is the reason the protocol exists.

Servers that never call a model

A server does not talk to a model, so the first five parts need no API key at all. The in-memory client you meet in the first part runs the real protocol against your server with no process and no port. Part 6 adds an agent that calls a real model, openai/gpt-oss-120b on Groq's free tier, so it needs a free Groq key.

Because a server never calls a model, you can build and test the whole shop server without a key. A model only enters the picture when a host, or your own agent, decides to call one of the tools the server lists.

Where you use MCP

  • Giving one team's tools to Claude Code, Claude Desktop and Cursor from a single server.
  • Exposing a database or an internal API to an agent through a stable, described interface.
  • Sharing prompts and reference documents with everyone on a team as slash commands and attachable resources.
  • Building your own agent that lists a server's tools and calls them in a loop.

Advantages and limits

  • Advantage: a tool is described once and every MCP host can list and call it.
  • Advantage: the SDK builds names, descriptions and JSON Schemas from your type hints.
  • Advantage: tools, resources and prompts, transports and human approval are part of the protocol.
  • Limit: for a function only your own code will ever call, plain Python is shorter.
  • Limit: MCP is stateless, so conversation memory lives in the host, not the server.

Prerequisites

  • Python 3.10 or later.
  • Functions, type hints, Pydantic and async from Python for AI. Part 5 uses FastAPI from APIs for AI.
  • No API key for the first five parts; the agent in part 6 needs a free Groq key, set up in the setup lesson.
Back toAll frameworks

Every expert started right here.