AI Agents Engineering · Model Context Protocol · Lesson 1 of 3
MCP primitives: tools, resources, prompts
MCP (Model Context Protocol) is an open standard for connecting models to tools and data through a uniform interface — instead of a bespoke integration per app. Created by Anthropic, it was donated to the Linux Foundation's agentic-AI umbrella in late 2025 and is now supported by every major host (Claude, ChatGPT, Gemini, Copilot, Cursor, VS Code). Think USB-C for AI tools.
- Tools: callable functions with JSON-schema inputs — model-controlled actions (query a DB, send a message).
- Resources: readable data the client can load into context (files, records, docs) — application-controlled, addressed by URI.
- Prompts: reusable, parameterized prompt templates the server offers — user-controlled (e.g. slash-commands).
Who controls whatThe classic framing: tools are model-controlled (the model decides to call them), resources are app-controlled (the client decides what to load), prompts are user-controlled (the user invokes them). Naming this distinction signals depth.
◆ Lock it in
- MCP = a standard interface for tools/data across apps ('USB-C for AI tools').
- Primitives: tools (actions), resources (data), prompts (templates).
- Control model: tools = model, resources = app, prompts = user.
Feynman drill — say it out loudName MCP's three primitives and say who 'controls' each.
Step 1 rate your confidence · Step 2 pick your answer
What are the three core primitives an MCP server exposes?
Step 1 — how sure are you?