What is Claude Code? Claude Code is Anthropic’s coding agent that carries out programming work in your own project folder. You give an instruction in plain language, and Claude Code reads your files, edits code, runs commands and commits the changes to git.
Where Claude in the chat gives you an answer that you copy across yourself, Claude Code does the work in your project folder. You install Claude Code with a single command on Windows, macOS or Linux, or you open it without installing anything in the desktop app or the browser.
Claude Code is not included in Claude’s free plan. You need Pro, Max, Ultra, Team, Enterprise or a Claude Console account with prepaid credits.
How does Claude Code work?
Claude Code works in a session that you start in your project’s folder. You describe the task, Claude Code finds the relevant files, proposes a change and, in the default mode (auto), carries it out, with checks running in the background. Only in Manual mode does Claude Code ask for permission before a file changes or a command runs.
-
StartClaude Code reads CLAUDE.md in your project folder.
-
Describe the taskYou give the instruction in plain language.
-
Find filesClaude Code looks up the relevant files itself.
-
Propose a changeIt prepares the edit or command.
-
Permission mode checkAuto mode continues; Manual mode asks for your approval.
-
Carry outThe change is made in your project folder.
The session’s permission mode determines which actions Claude Code carries out without asking. Since version 2.1.283, a session starts in auto mode by default. Press Shift+Tab during a session to switch between modes: first to Manual, where you approve each change individually, and then to acceptEdits and plan.
According to Anthropic’s documentation, Claude Code reads the CLAUDE.md file in your project’s root folder at the start of every session. That is where you record agreements such as coding standards, preferred libraries and a review checklist. Claude Code also keeps lessons from earlier sessions in a memory of its own.
You connect external sources through the Model Context Protocol (MCP), an open standard for connecting AI tools to data. That lets Claude Code read design documents in Google Drive, update tickets in Jira or fetch messages from Slack. You set up recurring work as a skill, and with hooks a command runs automatically before or after an action, for example formatting the code after every file change.
What can you do with Claude Code?
With Claude Code you can hand off programming work that would otherwise mean manual editing. Anthropic’s documentation lists seven types of task:
- Writing tests for untested code, running them and fixing the failures
- Building features that touch several files, from a description in plain language
- Tracking down bugs: you paste the error message and Claude Code looks for the cause in the codebase
- Creating commits, branches and pull requests with a descriptive explanation
- Connecting tools through MCP, such as Google Drive, Jira and Slack
- Scheduling recurring tasks, such as a weekly check for outdated packages
- Piping in log files or changed files for analysis or a review
Scheduled tasks run as routines in Anthropic’s cloud and keep going while your computer is off. Tasks that need access to your own files are scheduled in the desktop app on your own computer.
Outside software development, the typical use is small scripts. A marketing agency that merges the same Search Console exports into a report every month can have Claude Code write a script for it once, test it and put it in the project folder.
How is Claude Code different from Claude in the chat?
Claude Code and Claude in the chat differ in who does the work. In the chat, you copy the answer into your files yourself; Claude Code makes the change in your project folder directly. Both use the same Anthropic model family.
The table sets out the five differences that determine which of the two suits a task.
| Feature | Claude in the chat | Claude Code |
|---|---|---|
| Where you work | claude.ai, desktop app and mobile app | Terminal, editor, desktop app and browser |
| File access | Files you upload | The whole project folder, without uploading |
| Running commands | No | Yes, including git, and in Manual mode only after your approval |
| Free plan | Yes | No, from Pro or a Console account |
| Strong at | Writing, analysing and sparring | Building, testing and editing code |
For writing and analysis, the chat remains the shortest route. Claude Code pays off as soon as a task touches several files or runs code. A third route is the Claude API, which lets you call Claude from your own software. In my own work with Claude, I use the chat for content and strategy. On my VPS, I also run a workflow in which Claude analyses Search Console data through the API, without Claude Code.
What do you need to get started?
For Claude Code you need three things: an account with access to Claude Code, a computer that meets the system requirements and a project folder to work in.
A Pro, Max, Team or Enterprise plan gives access to Claude Code, as does a Claude Console account. The system requirements according to Anthropic:
- macOS 13.0 or later
- Windows 10 version 1809 or later, or Windows Server 2019
- Ubuntu 20.04, Debian 10 or Alpine Linux 3.19 and later
- At least 4 GB of RAM and an x64 or ARM64 processor
- An internet connection
Choose Team or Enterprise if you work with client code or client data. Those plans fall under Anthropic’s Commercial Terms, with a data processing agreement and no training on your conversations by default. Pro and Max fall under the Consumer Terms, where a privacy setting determines whether your conversations are used for training.
How do you install Claude Code on Windows and Mac?
To install Claude Code, you run a single command in the terminal. Anthropic recommends the native installer, because it updates Claude Code automatically in the background. The command shows no progress while it downloads; when it has finished, open a new terminal window before you continue.
How do you install it on a Mac or on Linux?
On a Mac and on Linux, you install Claude Code with this command in the Terminal:
curl -fsSL https://claude.ai/install.sh | bashWith Homebrew, you install it with brew install --cask claude-code. That version does not update itself, so run brew upgrade claude-code from time to time. For Debian, Fedora and Alpine, Anthropic also offers packages through apt, dnf and apk.
How do you install it on Windows?
On Windows, the install command depends on the window you are working in: PowerShell or the Command Prompt (CMD). You can recognise PowerShell by the PS before the prompt, for example PS C:\Users\Name>.
In PowerShell:
irm https://claude.ai/install.ps1 | iexIn CMD:
curl -fsSL https://claude.ai/install.cmd -o install.cmd && install.cmd && del install.cmdIf you get a message that ‘&&’ is not a valid statement separator, you are in PowerShell and need the first command. If you get a message that ‘irm’ is not recognised, you are in CMD.
You do not need administrator rights. Git for Windows is optional: without Git, Claude Code runs commands through PowerShell, with Git through Git Bash. With WinGet, you install it with winget install Anthropic.ClaudeCode; that installation does not update itself either.
How do you check that the installation works?
To check the installation, type claude --version in a new terminal window. A working installation shows a version number followed by (Claude Code).
A message that claude cannot be found means the installation folder is not yet in your PATH. claude doctor gives you an overview of the installation, your settings and any warnings with a suggested fix. claude update installs the latest version straight away, without waiting for the automatic update.
How to use Claude Code: your first session
To start using Claude Code, open the terminal in your project’s folder and type claude:
cd path/to/your-project
claudeThe first time you use it, Claude Code asks you to log in. The login runs through your browser, after which Claude Code remembers your details.
Start with questions that change nothing, so you can see how Claude Code reads your project:
- What does this project do?
- Which technologies does this project use?
- Explain the folder structure.
Then try a small change, for example: add input validation to the registration form. In the default auto mode, Claude Code makes the change straight away. If you want to review every change first, press Shift+Tab to switch to Manual, after which Claude Code asks for approval, which you give by choosing Yes.
Four commands cover everyday use:
/helpshows the available commands/clearclears the conversation/exitends the sessionclaude -ccontinues the last conversation in that folder from the terminal
Can you use Claude Code without a terminal?
You can also use Claude Code without a terminal: in the desktop app, in the browser and as an extension in your editor. All versions run on the same Claude Code, so your CLAUDE.md, settings and MCP connections work everywhere.
In the Claude desktop app, you open the Code tab. The app already includes Claude Code, so you do not need a separate installation. You can review changes visually, run several sessions side by side and schedule recurring tasks; a paid plan is required.
At claude.ai/code you start a session in the browser without a local installation, including for repositories that are not on your computer. A long-running task keeps going while you do something else, and you can review the result later in the browser or in the Claude app on iOS and Android.
The extension for VS Code, which also works in Cursor, shows changes directly in the code and lets you review a plan before Claude Code carries it out. The plugin for JetBrains editors such as IntelliJ IDEA and PyCharm requires a separate installation of Claude Code.
Which Claude plan do you need for Claude Code?
For Claude Code you need a paid plan: Pro, Max, Ultra, Team or Enterprise. Claude Code is not a separate add-on; it is included in those plans at no extra cost and uses the same usage limit as your chat conversations.
A long session in a large codebase uses more capacity than a conversation does, so how heavily you use it determines whether Pro is enough or you need Max. For the monthly cost of Pro, Max and Team, see the current Claude pricing.
The second route is a Claude Console account with prepaid credits. You then pay per token used instead of a fixed monthly amount. When you first log in, Claude Code creates a dedicated workspace in the Console, so you can track the costs separately.
Can you use Claude Code without programming experience?
You can use Claude Code without programming experience for small tasks you can check, but the terminal, project folders and git remain the ground it works on. Without a basic understanding of files and version control, you lose track of what is changing.
So work in a folder under git or on a copy, so that every change can be undone. From my time at major banks, I bring the four-eyes principle to AI work: read every proposed change and understand what it does before you approve it. For that way of working, press Shift+Tab to switch to Manual mode. Since version 2.1.283, a session starts in auto mode by default, and in that mode Claude Code makes changes without asking for approval first.
For work without code, such as processing documents or organising folders, Claude Cowork is a better fit. Cowork brings the same agentic approach to a visual environment, without a terminal.
You can also automate recurring work with Claude without learning to program first. In the AI Automation Course, you build an automation for a work process from your own practice, without code through Make or Zapier, or with Python.
Frequently asked questions about Claude Code
Daily use of Claude Code raises questions that go beyond installation and the first session.
Can you undo changes made by Claude Code?
Yes, you can undo changes made by Claude Code with /rewind or by pressing Esc twice when the input field is empty. Claude Code saves a checkpoint with every prompt, and in the menu you choose whether to restore only the code or the conversation as well. Files changed only through shell commands fall outside those checkpoints. Checkpoints are therefore no substitute for git.
How do you switch models in Claude Code?
You switch models in Claude Code with /model, followed by an alias such as sonnet or opus, or on its own to open a picker. At start-up you choose a model with claude –model, and you set a fixed choice in your settings file with the model field. According to Anthropic, /model also saves your choice as the default for new sessions.
Can you use Claude Code in a language other than English?
Yes, you can have Claude Code answer in another language by default by setting the language option in your settings file, for example to german or dutch. Claude Code passes that value on word for word as an instruction to always answer in that language. The same value also sets the language of automatic session titles and of voice dictation, where that language is supported. Commands such as /help and claude –version stay in English.
How do you uninstall Claude Code from your computer?
On a Mac or on Linux, after a native installation, you remove Claude Code with rm -f ~/.local/bin/claude and rm -rf ~/.local/share/claude. On Windows, you delete the same folders in your user profile using PowerShell, and with Homebrew or WinGet you use that package manager’s uninstall command. Your settings and session history are stored separately in ~/.claude and ~/.claude.json, which you only delete if you want to remove everything.
