Skip to main content

What Claude Code Is

Beginner

Claude Code is Anthropic's agentic coding tool. Unlike a chat window, it can actually do things in your project: read and edit files, run shell commands, search the codebase, and call external tools — all with your permission.

The mental model: an agentic loop​

This is the one idea that makes everything else make sense:

You give an objective in plain language ("add tests for the auth module and fix what fails"). Claude plans, acts, observes the result, and repeats until the goal is met. You stay in control via permissions and Plan Mode.

Where you can run it​

  • Terminal (CLI) — the original surface; works in any shell.
  • IDE extensions — VS Code and JetBrains, with inline diffs.
  • Desktop and web — and it shares your settings, hooks, and permissions across surfaces.

What you'll configure (in rough order of leverage)​

  1. CLAUDE.md — persistent project instructions. Highest impact, lowest effort.
  2. Plan Mode — investigate and propose before any edits run.
  3. Permissions — what Claude may do without asking.
  4. settings.json — the full config system.
  5. Slash commands, hooks, skills, subagents, MCP servers — power features, layered on as you need them.

Your first session (the shape of it)​

  1. Install and authenticate (see the official docs for current commands).
  2. cd into a project and start Claude Code.
  3. Run /init to generate a starter CLAUDE.md.
  4. Ask for something small and concrete: "Explain how routing works in this app."
  5. Then try a change in Plan Mode first, review the plan, and let it execute.

:::tip Start read-only For your first real task, use Plan Mode — Claude investigates and shows you a plan without touching files. It's the safest way to build trust. :::

Next​