The short answer: install with one command, run claude inside a project folder, sign in, ask it a question about the code, then give it one small change and a check it can run. That is the whole loop. Everything else, CLAUDE.md files, skills, hooks, subagents, is an add-on you bring in when a specific annoyance shows up, and the table below says which add-on fits which annoyance. You need a Claude Pro, Max, Team or Enterprise plan, a Console account, or a supported cloud provider; what each costs is on our Claude Code pricing page.
Install and first session
From Anthropic’s quickstart, checked 29 September 2026:
- Install. macOS, Linux or WSL:
curl -fsSL https://claude.ai/install.sh | bash. Windows PowerShell:irm https://claude.ai/install.ps1 | iex. Homebrew (brew install --cask claude-code) and WinGet (winget install Anthropic.ClaudeCode) also work, but do not auto-update. On native Windows, Git for Windows is recommended. - Check. Open a new terminal and run
claude --version. - Start.
cdinto your project and runclaude. You are prompted to log in on first use;/loginswitches accounts later. - Ask before you change. Try “what does this project do?” or “where is the main entry point?”. Claude reads files as needed; you do not paste context.
- Make one small change. “add input validation to the signup form”, then approve the edit if asked.
If you set an ANTHROPIC_API_KEY environment variable, Claude Code asks to use that key instead of your subscription, which bills per token.
It also runs in VS Code, JetBrains IDEs, a desktop app, the web at claude.ai/code and in CI, all sharing the same CLAUDE.md files, settings and MCP servers, per the overview.
The loop that works: check, plan, then code
Anthropic’s best-practices page is built on one constraint: the context window fills fast and performance degrades as it fills. Its main advice follows from that.
- Give it a check to run. A test suite, a build, a linter or a screenshot to compare. Without one, “looks done” is the only signal and you become the verification loop.
- Explore, plan, then implement. Press
Shift+Tabuntil the status bar shows plan mode, so Claude reads without editing, then approve the plan. Anthropic says to skip planning if you could describe the diff in one sentence. - Be specific. “fix the login bug where users see a blank screen after entering wrong credentials” beats “fix the bug”.
- Clear between unrelated tasks.
/clearresets context. If you have corrected the same issue twice, Anthropic recommends clearing and restarting with a better prompt. - Undo is built in.
Esctwice or/rewindrestores earlier conversation and code state, though it does not track changes made by shell commands, so keep using git.
The default for permissions matters here. From version 2.1.283 the docs describe auto mode, where a second model reviews actions instead of you, as the starting mode for interactive terminal and VS Code sessions; on earlier versions it is the starting mode only on Pro, Max and Team plans, and elsewhere Claude asks before edits. Shift+Tab cycles modes either way, and our Codex comparison sets this against OpenAI’s defaults.
Which extension to add, and when
This is the part of the docs people skip. Anthropic’s extension page pairs each feature with a trigger. Condensed:
| You notice… | Add | Note from the docs |
|---|---|---|
| Claude gets a convention or command wrong twice | CLAUDE.md | Loaded every session; keep under about 200 lines. /init writes a starter |
| You keep retyping the same prompt or playbook | A skill | Loads on demand; use /name to invoke |
| Claude cannot see data in some other system | An MCP server | claude mcp add ...; tool schemas load only when used |
| A side task floods the chat with output | A subagent | Runs in its own context and returns a summary |
| Something must happen every time, such as linting after edits | A hook | Deterministic; a CLAUDE.md line is only a request |
| A second repo needs the same setup | A plugin | Bundles skills, hooks, subagents and MCP |
The one distinction worth memorising: instructions in CLAUDE.md or a skill are advisory, hooks are guaranteed. If a rule must hold every time (“never edit .env”), the docs say make it a hook. How to write the file itself is in CLAUDE.md best practices, and keeping long builds from degrading is in protecting your context window.
For a fuller comparison of the four, see hooks, subagents, skills or CLAUDE.md.
Connecting outside services is covered in Claude Code MCP servers.
Running it without watching
claude -p "your prompt" runs once and exits, for scripts and CI; claude -c continues your last conversation; claude --permission-mode plan starts in plan mode. Anthropic’s docs also describe parallel sessions in git worktrees and a fan-out pattern that loops claude -p over a file list with --allowedTools to limit what it may do unattended. Try any of these on two or three files before running the lot.
Where to go next
The tool itself is assessed in the Claude Code review. Against an editor, see Claude Code vs Cursor; against a suite of assistants, the best AI code editor guide.
Method
Researched from Anthropic’s Claude Code documentation on 29 September 2026. We have not run a controlled test of Claude Code for this page. Commands and version notes are as Anthropic states them and change often; Anthropic’s docs are the reference if anything here differs. Re-checked every 60 days.
Common questions
Do I need a paid plan to use Claude Code?
You need a Claude Pro, Max, Team or Enterprise plan, a Console account with API credit, or a supported cloud provider such as Bedrock. There is no free tier.
Do I need to use the terminal?
No. It also runs in VS Code, JetBrains, a desktop app and the web, though the terminal is the full-featured version per the docs.
What should I do first in a new project?
Ask it to explain the codebase, then run /init to generate a starter CLAUDE.md and trim it by hand.