Claude Code: a beginner's guide
Read in : Français

Contents
- What is Claude Code and how is it different?
- Which plan do you need for Claude Code?
- Install Claude Code in three minutes
- Your first session: the right questions in the right order
- CLAUDE.md: your project's memory
- Permission modes: who decides, you or the agent?
- Undo a mistake with /rewind
- Pick the right model
- Six beginner mistakes to avoid
- Frequently asked questions
- Frequently asked questions
- Sources
Claude Code is Anthropic's coding agent: you describe what you want in plain English, and it reads your project, edits the files, runs the commands and checks the result. This Claude Code guide takes you from installation to your first feature, with the settings that prevent nasty surprises. All you need is a paid Claude plan and a project folder.
What is Claude Code and how is it different?
Claude Code is an agent, not just a chat: it acts inside your project instead of handing you code to copy and paste. The official docs describe it as a tool that "reads your codebase, edits files, runs commands, and integrates with your development tools".
Unlike an editor such as Cursor, Claude Code was born in the terminal. It now runs on several surfaces:
| Surface | Best for | Worth knowing |
|---|---|---|
| Terminal (CLI) | People who want every feature | The most complete version |
| VS Code (and Cursor) | People who want diffs inside the editor | Official extension with inline diffs and plan review |
| JetBrains | IntelliJ, PyCharm, WebStorm users | Plugin that relies on the CLI, installed separately |
| Desktop app | People who dislike the terminal | "Code" tab in the Claude app, paid plan required |
| Web and mobile | Long tasks with no local setup | At claude.ai/code, from a GitHub repository |
If you are new and the terminal scares you, the desktop app is the gentlest way in. Everything in this guide works the same way there.
Which plan do you need for Claude Code?
Claude Code requires a paid Claude plan or an API account; the free Claude plan does not include it. Prices checked on November 21, 2026 on the official Claude pricing page:
| Plan | Price | Usage | Claude Code |
|---|---|---|---|
| Free | $0 | Limited | Not included |
| Pro | $20/month, or $17/month billed annually ($200 up front) | Baseline | Included |
| Max | $100/month | 5x Pro usage | Included |
| Ultra | $200/month | 10x Pro usage per week, no 5-hour limit | Included |
| Team | $25 per standard seat ($20 annual), $125 per premium seat ($100 annual) | Depends on seat | Included |
| Console (API) | Pay as you go, prepaid credits | Depends on spend | Included |
Prices in US dollars, before tax. For a first project, Pro is enough to learn. If you keep hitting the limit, Max starts to pay for itself. The cost documentation gives an enterprise benchmark: on average around $13 per developer per active day, $150 to $250 per developer per month, and under $30 per active day for 90% of users. Those are heavy professional workloads, not a weekend side project.
Install Claude Code in three minutes
Installation is a single command, and Claude Code asks you to sign in the first time you run it.
- Open a terminal. On Mac: the Terminal app. On Windows: PowerShell.
- Run the official installer.
# macOS, Linux, WSL
curl -fsSL https://claude.ai/install.sh | bash
# Windows PowerShell
irm https://claude.ai/install.ps1 | iex
- Check it: open a new terminal and type
claude --version. You should see a version number. - Start Claude Code in your project:
cd my-project
claude
- Sign in with your Claude account in the browser window that opens.
Alternatives: brew install --cask claude-code on Mac, winget install Anthropic.ClaudeCode on Windows. Note that the docs say those two methods do not auto-update, unlike the native installer. On native Windows, install Git for Windows so Claude Code can use Bash.
Your first session: the right questions in the right order
Get Claude Code to explain your project before you ask it to change anything. The first prompts suggested in the docs are simple on purpose:
what does this project do?
explain the folder structure
where is the main entry point?
Then try one tiny change to see how it asks for permission:
Add a footer with the current year to the home page. Show me the change before applying it.
You don't need to paste files into the chat: Claude Code reads what it needs on its own. Type /help to list commands and /resume to pick up an earlier conversation.
CLAUDE.md: your project's memory
A CLAUDE.md file holds the instructions Claude Code reads at the start of every session. Each session starts with an empty context window: without this file, you re-explain everything every time.
- Where it goes:
./CLAUDE.mdat the project root (shared with your team through Git),~/.claude/CLAUDE.mdfor preferences across all your projects,./CLAUDE.local.mdfor personal project settings (add it to.gitignore). - How to create it: run
/init. Claude analyzes the project and generates a first file with the build commands, test instructions and conventions it finds. - If your project already has an AGENTS.md (the format read by Codex and other agents), Claude Code can read it too.
Here is a template to adapt for a vibe coding project:
# Project: booking app (Next.js + Supabase)
## Commands
- Run locally: npm run dev
- Check before handing back: npm run build && npm test
## Rules
- Explain your plan in 3 lines before any change touching more than 2 files.
- No secret keys in front-end code; secrets go in .env.local.
- Every new Supabase table gets RLS policies enabled.
- Commit to Git after every step that works.
The two middle rules are worth their weight in gold: they prevent API keys leaking into the front end and the Supabase RLS mistake. For more on these context files, read our guide to CLAUDE.md, AGENTS.md and rules files.
Permission modes: who decides, you or the agent?
The permission mode sets what Claude Code may do without asking you. Switch modes at any time with Shift+Tab.
| Mode | What runs without asking | Best for |
|---|---|---|
default (manual) | File reads | Unfamiliar code, sensitive work |
acceptEdits | Reads, file edits, common file commands | Iterating on code you are reviewing |
plan | Reads and exploration, no edits until you approve | Exploring and planning before changes |
auto | Everything, with a second model checking each action in the background | Long tasks, fewer interruptions |
dontAsk | Only pre-approved tools | Scripts and CI |
bypassPermissions | Everything, no checks | Isolated containers and VMs only |
A good beginner routine: plan mode for any feature that touches several files, read the plan, then approve. It is the same logic as writing the spec before the prompt. bypassPermissions has no business on your personal laptop.
Undo a mistake with /rewind
Claude Code saves a restore point before every prompt you send. Type /rewind, or press Esc twice when the input is empty, to open the rewind menu. Then choose to restore the code, the conversation, or both.
The docs say Claude Code keeps file snapshots for the 100 most recent checkpoints in a session. It is a safety net, not version control: keep committing with Git, even when vibe coding.
Pick the right model
The /model command lets you choose the model. Aliases save you from remembering version numbers:
sonnetfor everyday coding;opusfor complex reasoning;haikufor quick, simple tasks;opusplanuses Opus while planning, then Sonnet to execute.
A more powerful model burns through your allowance faster. To compare model families, see which AI model is best for coding.
Six beginner mistakes to avoid
- Starting Claude Code in your home folder instead of the project folder. It works wherever you launch it.
- Asking for a whole app in one message. Break it into verifiable steps, as shown in the 5-step prompting method.
- Never reading diffs. At least skim the changed files and read error messages.
- No commits between steps. A step that works deserves a commit.
- Forgetting to ask for verification. End requests with "run the build and tests, fix any errors". To test without being a developer, read our dedicated guide.
- Shipping without checks. Go through the pre-launch security checklist first.
To compare Claude Code with today's other tools, our roundup of the best vibe coding tools and our Cursor vs Claude Code vs Codex face-off will help you decide.
Frequently asked questions
Frequently asked questions
Is Claude Code free?
No. Claude Code is not part of the free Claude plan. You need at least the Pro plan, at $20 a month or $17 a month billed annually as of November 21, 2026, or a Claude Console account billed per use through the API. The Max, Ultra, Team and Enterprise plans include it too, with more usage.
Do I need to know the terminal to use Claude Code?
You only need to open a terminal, move into a folder with cd and type claude. The docs even include a terminal guide for complete beginners. If you would rather skip the command line entirely, the Claude desktop app includes Claude Code in a "Code" tab, and the VS Code extension works inside the editor.
What is the difference between Claude Code and Claude on claude.ai?
Claude on claude.ai is a conversational assistant: it answers you and suggests code. Claude Code is an agent: it works directly in your project files, runs commands, executes tests and fixes its own errors. For building and maintaining a real application, Claude Code removes the copy-paste back-and-forth.
Can Claude Code break my project?
Like any agent, it can make a change you didn't want. Three safeguards reduce the risk: the permission mode (stay on default or plan at first), the /rewind command that restores code to before a prompt, and Git, which keeps your full history. Never use bypassPermissions outside an isolated environment.
Does Claude Code work in languages other than English?
Yes. You can write requests and your CLAUDE.md in French, Spanish or any major language, and Claude replies in kind. Commands such as /init, /rewind and /model, and file names, stay in English. For code shared with an international team, ask for English comments and commit messages in your CLAUDE.md.


