Claude Code
by Anthropic
Getting-started guide
Anthropic's coding agent: point it at a codebase, describe the change, and it reads, edits and runs things.
Current prices: Pro $17/mo with annual ($20 billed monthly) · See all plans
Last reviewed 2026-09-07 · for Claude Code
You open it inside a project folder on your own computer and describe the change you want in plain English. It reads the files itself, proposes edits, runs commands to check its own work, and asks permission as it goes. You read what it did and say what to fix.
Claude Code is Anthropic's coding agent — AI that takes actions on its own rather than only answering. Instead of handing you a snippet to paste, it reads, edits and runs a project's files on your machine, across a whole codebase rather than one file.
It runs in a terminal (a text window where you type instructions instead of clicking), in editors like VS Code and JetBrains, and in desktop, browser and phone apps.
Cursor is an editor you work inside; Claude Code is something you hand a task to and check on. GitHub Copilot finishes the line you are typing; this takes the whole change.
This is a developer tool, and the entry requirement is a project of code on your own disk that you already work on. Never used a terminal or version control? A browser assistant like ChatGPT writes and explains code with nothing to install, and is the better place to begin.
You need a paid Claude plan — currently Pro, Max, Team or Enterprise — or an Anthropic Console account billed for what you use. Pro is the entry plan and Max costs more for more usage; current prices are on its Launchpad page.
Usage runs on rolling windows rather than a credit balance: a limit resetting roughly every five hours, plus a weekly ceiling. Anthropic describes Pro as suited to light work on small repositories. Hit the ceiling and you wait, or turn on paid usage credits. /usage shows where you are, and long sessions cost more, because each turn resends the whole conversation.
Official source: Anthropic's official Claude Code documentation
Install only from Anthropic's official installer or package listings — lookalikes exist, and the steps change often enough that blog-post instructions are a bad habit.
Start on a project tracked in version control with nothing unsaved: that tracking is how you see every line it changed, and how you undo it.
Permission prompts offer allow once and allow from now on. The second outlives the task, so grant narrowly and audit with /permissions.
On Pro, Max and Team, sessions currently start in auto mode, where a safety model reviews actions instead of you; Shift+Tab switches to manual mode, read-only until it asks. The flag that skips checks entirely is for throwaway containers.
Your code is sent to Anthropic's models to be processed — how the tool works, not a side effect. On a client's private codebase, make that choice deliberately, and check your contract first.
Walk me through this codebase. What does it do, where does a request enter, and which three files should I read first?
Why this one: It reads the actual files instead of reasoning from what you pasted, and manual mode lets you learn the permission prompts while it can change nothing.
Add input validation to the login handler, then run the project's tests and fix anything that fails.
Why this one: It is judged on the loop, not the snippet — edit, run, see the failure, fix. A pass-or-fail task is the only way to see whether that loop closes on your project.
Creates a plain text file named CLAUDE.md in the project's top folder, holding the test command and the rules it should follow every session.
Why this one: Claude Code reads that file from the project root at the start of every session, so your test command and never-touch-this rules apply without retyping. Keep it short: a long one eats into every session, relevant or not.
Want the full picture first? See the Claude Code overview for pricing, the free tier and related tools.
