Installing Claude Code is the least interesting part of this course, so let us make it short and un-scary.
What the install actually does
It puts one program on your machine, called claude, into a folder that your shell already searches. That is the whole story.
It does not run in the background. It does not watch your files. It does nothing at all until you type claude and press Enter — and then it only sees the folder you started it in.
Get the command from the source
Install instructions change as the tool evolves, and a wrong one wastes an evening. Open Anthropic’s official install page and copy the current command for your operating system from there. That page is right; this page is old the day it is written.
What you will find there is roughly: an install script for macOS and Linux, and package-manager options on macOS, Windows and common Linux distributions. On Windows, running it inside WSL is a well-trodden path if the native option gives you trouble.
Try the PATH idea for yourself
Your shell does not search your whole disk when you type a command. It looks in a specific list of folders, in order. That list is the PATH variable, and you can print it.
Run echo $PATH in the terminal below, then start the agent with claude.
Quick check
You type `claude` and get `command not found`. What is the first thing to try?
The first run: signing in
The first time you start it, Claude Code needs to know how you are paying for the model. There are two normal answers:
- A Claude subscription plan. You sign in through the browser, and your usage counts against the plan.
- An API key from the Anthropic Console. You are billed per token instead.
Both work. Which one is better for you depends on how much you will use it, and the current pricing lives on Anthropic’s pricing page — not here, where it would be out of date within weeks.
Keys are secrets
If you go the API-key route, that key is a password with a credit card attached. It belongs in your environment or a that Git ignores — never pasted into a source file, and never committed.
Two commands worth knowing on day one
Claude Code ships with a diagnostic command that checks your installation and exits, and an update command that pulls the newest version. When something feels wrong before you have written a single prompt, the diagnostic is the right first move — run claude --help to see the exact names in your version, since the surface changes over time.
That habit — --help first, tutorial second — will serve you for the rest of this course.