MCP (Model Context Protocol) is an open standard, created by Anthropic and released in November 2024, that lets AI applications such as Claude Desktop and Cursor connect to external tools, files, and data sources through one shared interface instead of custom code for each connection. It works through a Host, Client, and Server.
What you will be able to do
- Explain what MCP is and why it replaces custom, one-off integrations between AI apps and tools.
- Describe what the Host, Client, and Server each do when an AI model requests a tool.
- Identify MCP's three building blocks: tools (actions), resources (data), and prompts (templates).
- Tell someone how MCP differs from a regular API when they ask.
- Work out why MCP changes integration effort from M x N connections to M + N.
Before you start
- No coding background required, MCP is meant for anyone who wants AI connected to other services, not just developers.
- Access to an MCP-capable AI application, such as Claude Desktop or Cursor, since many already support MCP out of the box.
- Basic familiarity with how you currently interact with an AI assistant, such as a chat window, IDE, or voice app.
Reference
| Piece | Category | What it does | Example from the article |
|---|---|---|---|
| Host | Architecture | Where you interact with the AI model | Chat window, IDE like Cursor, voice app |
| Client | Architecture | Translator inside the Host that converts the model's request into MCP language and talks to the Server | Turns "I need to check the weather" into an MCP request |
| Server | Architecture | Toolbox holding the real tools, waits for any MCP-speaking app to ask for them | One server holding get_weather, get_forecast, get_air_quality |
| Tools | Primitive | Actions that perform or change something | Download a file, run code, send an email |
| Resources | Primitive | Data the AI model can read but not modify | Company handbook, spreadsheet, weather data |
| Prompts | Primitive | Instruction sets or templates that help start a task | A "Code Review Mode" prompt that reminds the model what to check |
Common errors and fixes
| What goes wrong | The fix |
|---|---|
| Worrying the AI model could run a dangerous command through MCP | Good MCP applications ask your permission first before running anything risky, you stay in control of execution |
| Assuming MCP is just another name for an API | An API is one specific connection between two systems, MCP is a shared protocol so you build one MCP connection instead of a separate API integration per tool |
| Assuming MCP is only for developers | MCP is for anyone who wants AI to connect with other services, many AI applications already support pre-built MCP servers with no code required |
| Not knowing where the actual tools live | Tools live on MCP Servers, which can run on your own machine or be hosted online |
Read the full walkthrough
The complete lesson, with screenshots and any downloads, is published on Substack as part of MCP Masterclass: Connect AI to Everything.
More in this section
Continue the course
Browse all lessons in the MCP Masterclass: Connect AI to Everything course, or subscribe to the GenAI Unplugged newsletter to get new lessons in your inbox.
