context-course/pr_2/en/unit2/introduction.md

Introduction to Model Context Protocol

Model Context Protocol (MCP) is an open standard for letting AI applications interact with external data and tools. It sits between your agent and everything else: files, databases, APIs, services.

Unit 1 covered skills which are static, pre-written context that lives alongside your project. Skills are great for fixed knowledge, but they can't fetch new information or take live actions. MCP fills that gap by giving an agent dynamic context: it can call functions, read files, query databases, and interact with APIs through a single protocol.

The M×N Problem

Imagine you want to build an AI agent that can:

• Read files from your local system
• Query your company's database
• Call Slack APIs
• Access Jira tickets
• Search a knowledge base

Without a standard protocol, you'd need to write custom integration code for every combination of agent + data source. That's an M×N problem: N agents × M data sources = N×M custom integrations.

MCP eliminates this. You write your data sources as MCP servers once, and they work with any MCP-compatible agent. One adapter, infinite reuse.

MCP as Universal Adapter

MCP uses a simple architecture:

• Host: Your AI application (Claude, VS Code, an agent framework)
• Client: Protocol handler inside the host
• Server: Your external program exposing capabilities

The client and server communicate using JSON-RPC 2.0 messages. You write servers; the host handles the protocol.

Skills + MCP

It's important to mention that MCP was built before skills and in many ways, skills have consumed some of MCP's functionality. However, MCP is still a powerful tool for extending your agent's capabilities, especially under authentication scenarios.

Skills are perfect for teaching an AI how to do something (write prompts, analyze emails, code style guidelines).

MCP is perfect for giving an AI the ability to do something (read files, call APIs, query databases).

You'll often use both: Skills provide the knowledge, MCP provides the tools and data.

What You'll Build

This unit covers MCP architecture (hosts, clients, servers, and the three capability types), building servers with FastMCP and Gradio, configuring agents as MCP clients, deploying to Hugging Face Spaces, and a hands-on project tying it together.