Core Concepts of Claude Code
Claude Code has 7 key building blocks: Tools, Permissions, Hooks, CLAUDE.md, MCP, Memory, and Skills. Together they make Claude Code powerful yet safe.
The Big Picture
Here is how the 7 core concepts relate to each other. Claude Code sits in the center, and everything else plugs into it.
Explain Like I'm 12
Cheat Sheet
| Concept | One-liner | Link |
|---|---|---|
| Tools | Built-in actions: read files, edit code, run commands | CLI Basics |
| Permissions | You control what Claude can do automatically | Tools & Permissions |
| Hooks | Shell commands that auto-run on events | Hooks |
| CLAUDE.md | Project instructions Claude reads every session | CLAUDE.md |
| MCP | Protocol for connecting external tools | MCP Servers |
| Memory | Persistent notes across sessions | Memory System |
| Skills | Reusable prompt templates for workflows | Skills |
The 7 Building Blocks
Let's walk through each concept. Think of these as the moving parts that make Claude Code tick.
Tools
Tools are the things Claude Code can actually do. Read a file, edit some code, run a shell command, search your codebase — each of these is a tool. Without tools, Claude would just be talking. With tools, it can act on your project.
Deep dive →Permissions
You wouldn't give a new employee the keys to everything on day one. Permissions work the same way. They control which tools Claude can use automatically and which ones need your approval first. You stay in control.
Deep dive →Hooks
Hooks are shell commands that fire automatically before or after Claude uses a tool. Want to auto-lint every file Claude edits? That's a hook. Want to block certain commands? Also a hook. They're your guardrails and automations.
Deep dive →CLAUDE.md
Every time Claude Code starts a session, it reads your CLAUDE.md file. This is where you tell Claude about your project: what language you use, what conventions to follow, what to avoid. Think of it as a briefing document.
Deep dive →MCP (Model Context Protocol)
MCP is how Claude Code connects to the outside world. Need to query a database, hit an API, or use a third-party service? MCP is the standardized protocol that makes that possible without custom integration code.
Deep dive →Memory
Claude Code forgets everything between sessions — unless you use Memory. Memory lets Claude persist notes about your preferences, your project, and things it learned. Next session, it picks up where it left off.
Deep dive →Skills
Skills are reusable prompt templates saved as SKILL.md files. You trigger them with slash commands. Got a workflow you repeat often — like creating a new component or running a deploy checklist? Turn it into a Skill and never retype it.
Deep dive →Test Yourself
Q: Name 3 of Claude Code's 7 key concepts.
Q: What controls what Claude can do automatically?
Q: Where does Claude read project instructions from?
Q: What connects Claude Code to external services?
Q: What are Skills?