Lesson 3 · 8 min

MCP Overview

What the Model Context Protocol is and when to use it.

MCP is a JSON-RPC protocol for exposing tools, prompts, and resources to LLM clients. Build an MCP server when you want the same toolset reusable across Claude Code, the desktop app, and your own agents.

Production scenario

Real-world example: Internal Notion MCP server for engineering

An engineering org runs a tiny MCP server that exposes their Notion workspace:

Tools: search_pages(query), create_page(parent, title, content)
Resources: notion://page/{id} for any page, addressable and cacheable
Prompts: /notion-summarize-meeting template the team uses every Monday

The same server is wired into Claude Code (engineers use it from the terminal) and the desktop app (PMs use it from a normal chat window). One implementation, two surfaces.

Why this matters: MCP's value isn't theoretical — it's the moment you realize the toolset you built for one client is now reusable across every client that speaks the protocol.

Knowledge points in this lesson

MCP = Model Context Protocol
It's JSON-RPC for LLM tools and data
Three primitives: tools, resources, prompts
stdio for local, HTTPS for remote
Same server reusable across LLM clients
Auth lives at the transport layer

Quick check

Tool Design & MCPSelect one

Should you expose a slow-changing reference dataset as an MCP resource or as a tool?