Original Pages

Free guide · Claude Code

Getting started with Claude Code

Level · Start here

This page gets you from a blank terminal to a first useful change with Claude Code. Install, sign in, open a real project, ask a read-only question, then make one small edit — before you touch CLAUDE.md or MCP.

Already have Claude Code installed?

  1. cd into a git project and run claude — first session.
  2. Ask it to explain the codebase before you ask it to edit — first question.
  3. Prefer a small, reviewable change over a multi-file rewrite on day one — first change.
  4. When the same correction shows up twice, put it in CLAUDE.md — CLAUDE.md that actually works.

Brand new? Ignore the box and start with install below.

Never used Claude Code? Start here

What Claude Code actually is

Claude Code is Anthropic’s agentic coding tool. The path most people mean is the terminal CLI: you open a project folder, run claude, and talk to an agent that can read files, edit them, and run commands — with your approval when it needs to do something consequential.

It is not the Claude model picker inside Cursor, and it is not the same product as Grok Bot. You can run Claude Code next to Cursor (in Cursor’s terminal, or via Anthropic’s editor integrations), but the two tools keep separate rules, MCP config, and settings except for CLAUDE.md, which both can read — see Using Claude with Cursor.

Official docs also cover Claude Code on the web, as a desktop app, in VS Code and JetBrains, in Slack, and in CI. This page sticks to the terminal CLI — that is the install path the Anthropic quickstart leads with, and the one the rest of Original Pages assumes when it talks about CLAUDE.md and /context.

Install

You need a terminal, a code project, and access via a Claude subscription (Pro, Max, Team, or Enterprise), a Claude Console account, or a supported cloud provider. Native install is what Anthropic currently recommends.

macOS, Linux, and WSL (native):

curl -fsSL https://claude.ai/install.sh | bash

Windows PowerShell (native) — your prompt shows PS C:\:

irm https://claude.ai/install.ps1 | iex

Windows CMD — your prompt shows C:\ without the PS:

curl -fsSL https://claude.ai/install.cmd -o install.cmd && install.cmd && del install.cmd

Mixing those two Windows shells is the classic first failure. If && is “not a valid statement separator,” you are in PowerShell — use the irm line. If irm is not recognized, you are in CMD — use the curl line. On native Windows, Anthropic recommends Git for Windows so the Bash tool works; without it, Claude Code falls back to PowerShell as the shell tool. WSL does not need Git for Windows.

Alternatives documented by Anthropic: brew install --cask claude-code on macOS (stable channel; there is also claude-code@latest), and WinGet on Windows. Native installs update in the background.

brew install --cask claude-code

If install fails with a curl 403, a syntax error near <, or another fetch error, use Anthropic’s Troubleshoot installation notes on the official quickstart rather than guessing a mirror.

Sign in

After install, run claude and complete login when prompted. Use the account path you already have access through — Claude.ai subscription, Console, or your organisation’s cloud provider setup. There is no separate “Claude Code only” password beyond that account flow.

Start your first session

  1. Open a terminal and cd into a real project folder — ideally one with git initialised.
  2. Run claude.
  3. Commit or stash anything precious before you let it edit. Session undo is not a substitute for git.

Claude Code works from the directory you launched it in. Starting it from your home folder and pointing at a nested project with vague prompts is how you get confident edits in the wrong tree.

Ask a read-oriented first question

Official onboarding leads with orientation before edits. Paste something like:

Explain this codebase. Point me to the main entry points, key modules, and anything I should read before making changes.

You are calibrating: how much of the repo it finds, where it invents structure, and whether you trust it enough for a small change. If you only want answers with no edits, say so explicitly in the prompt until you know the current permission prompts on your install.

Make one small change

Ask for a few safe improvements, pick one, and have it implement that one. Good first tasks are low risk: copy, a small UI fix, a missing test name. Then:

  1. Read the diff — do not accept a bundle unread.
  2. Run the checks your project already uses (tests, typecheck, lint, or a local build).
  3. Only after that, try a multi-file feature. Prefer an explicit plan in the conversation for anything that spans several modules.

Essential commands worth learning early

CommandWhat it is for
/helpBuilt-in help for the CLI you actually have installed.
/clearStart a fresh conversation when the thread is full of a failed approach.
/compactCompress context when the session is long but still useful.
/contextSee what loaded — including memory files — instead of guessing.

Exact slash-command names can grow with releases. If something here does not match your build, trust /help on your machine over a screenshot from a blog post.

Beginner mistakes worth skipping

  • Installing, then configuring forever. One small edit teaches more than an empty CLAUDE.md you wrote in the abstract.
  • Windows shell mix-ups. PowerShell and CMD need different install one-liners — see above.
  • Vague “fix the bug” with no file or error. Paste the stack trace or point at the path.
  • Skipping git. You want a one-command escape hatch before ambitious agent work.
  • Assuming Cursor rules apply here. They do not at runtime — CLAUDE.md vs .cursor/rules.

What to read next

When you are ready to make memory durable, read CLAUDE.md that actually works. If you also use Cursor day to day, Using Claude with Cursor covers which tool owns which file. For Amazon Bedrock specifically, see AWS integration.