For a Claude Code install CLI users can follow the official setup commands

What is the Claude Code CLI, and why use it in the terminal?#

The terminal scares almost everyone who gets into vibe coding without a software development background. It looks like a black window from hacker movies, but in reality it's just a line where you type commands. Claude Code needs it precisely so it can open your files, change them, and run checks by itself while you describe the task in words.

Anthropic describes the product this way.

The difference from a familiar chat is simple. In a browser, you copy code by hand, back and forth between the window and your editor. Claude Code goes into the project folder itself, reads the files it needs, makes the change, and shows you what it changed. You remain the one who decides and approves; it does the mechanical work for you.

If you really don't want to use the terminal, Anthropic has a desktop app with a regular window and a web version at claude.ai/code that runs right in the browser. But the most powerful and flexible option is the terminal version, which is what this whole article is about.

Where is the official Claude Code download, and how does installation work?#

The first point of confusion for beginners is where the official Claude Code website is and what to click to download it. Here's a table of all the official Anthropic addresses, so you don't have to guess.

You don't download an installer from the website as you would for a regular program, which goes against what people expect. You install it with a terminal command, which then downloads the right version for your system. This article has the exact commands for Mac, Windows, and Linux.

One common trap is the name: people often misspell it as clode code, clade code, слод код, or clode design. The product is called Claude Code, with au. Keep that in mind when looking for the official website: domains with similar spellings belong to someone else.

It's also worth knowing the exact npm package name. The canonical package is @anthropic-ai/claude-code, with the @anthropic-ai/ prefix. Anthropic itself holds the short name claude-code without the prefix—it is their official pointer to the same package. The simple rule: the publisher should be Anthropic, and the name to use for installation has the prefix.

For a Claude Code install Ubuntu and Debian have which system requirements?#

Claude Code's requirements are modest: Anthropic's servers do the heavy work; your machine only sends requests and receives responses.

A separate note about Node.js, since it's surrounded by the most myths. The native installer—the one Anthropic itself recommends—doesn't need Node.js at all: it installs as a single standalone file. Node.js is only required if you install Claude Code through npm, and then you need version 22 or later. Older articles say “Node 18 and above”—that's outdated; use 22+ as your guide.

You need a terminal for the main version. If you don't want to use one, install the desktop app with its regular window or use the web version at claude.ai/code. The terminal version is still more powerful in terms of features, so the rest of this article explains how to install that one.

For a Claude Code install macOS offers the native installer and Homebrew#

Open Terminal (it comes with every Mac: press Cmd+Space in Spotlight, type “Terminal”) and paste this one line:

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

The command downloads the installer from Anthropic's official address and installs Claude Code. Node.js isn't needed. The native method is also convenient because of its updates.

If you already use Homebrew, there's a second option:

brew install --cask claude-code

The difference between them is how updates work: the native installer updates itself, while Homebrew doesn't—you'll need to update it manually with the brew upgrade claude-code command. The program itself is the same, so choose whichever method you're used to. If you're not sure, use the first one.

For a Claude Code install Windows PowerShell provides which native command?#

First, figure out where you are. Windows has two shells, and each has its own command. Open a window and look at the start of the prompt: PS C:\> means PowerShell, while just C:\> without PS means Command Prompt (CMD).

For PowerShell (it's easiest to open it by right-clicking the Start button and choosing “Terminal”):

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

For Command Prompt (CMD):

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

There's a third option, using the built-in WinGet package manager:

winget install Anthropic.ClaudeCode

If you already have a working environment in WSL (Linux inside Windows), install Claude Code right inside WSL using the same command as on Linux. There's a separate WSL wrinkle involving Node, which I'll cover below in the common errors section.

For a Claude Code install Linux supports the native installer and package repositories#

The basic method for any popular distribution is the same standalone installer:

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

Node.js isn't required; the program comes as a single file and updates itself from then on.

If you're used to installing programs only through the system package manager, you can add Anthropic's official repositories: apt for Debian and Ubuntu, dnf for Fedora and RHEL, and apk for Alpine. They're signed with a key and have stable and latest channels. That's overkill for a beginner, so I'm leaving the details to the documentation in the sources section and recommending the command above by default.

There can be one snag on inexpensive servers. If the machine has very little free memory, installation may stop with the word Killed (code 137)—the system killed the process because it ran out of memory. The fix is to add a swap file: Claude Code needs about 512 MB of free memory to install.

How do you launch Claude Code for the first time?#

The first launch takes a couple of minutes. Here's the process.

1. Go to the project folder

   In the terminal, use the cd path-to-project command to go to the folder you want to work with, then type claude.

2. Sign in to your account

   A browser will open with a sign-in window. If it doesn't open automatically, press c to copy the link and paste it into a browser manually. If the browser displays a code (this can happen in WSL and on remote servers), paste that code back into the terminal.

3. Wait for confirmation

   Once you're signed in, the terminal will display Login successful and ask you to press Enter.

4. Ask your first question

   Ask something simple about the project, such as “what does this project do?” Claude will read the files and answer.

5. Approve the first change

   When you ask it to change something, Claude will show you the proposed changes and ask for confirmation. You can switch access modes with Shift+Tab.

It's worth learning a few basic commands inside the session right away: /help lists the commands, /clear clears the conversation history, and /exit exits (double Ctrl+D does the same). Claude will show you all of this itself if you type /help.

Sign in with a subscription or an API key: which should you choose?#

This is the first question that comes up when signing in.

There are two ways to sign in, and they differ in how you pay.

To sign in with a subscription, simply use your Claude.ai account. To sign in with an API key, get a key from Claude Console (platform.claude.com); charges are deducted from your prepaid balance as you work. Set the key in the ANTHROPIC_API_KEY environment variable, and Claude Code won't open the browser every time.

You can switch accounts within a session with /login and sign out with /logout.

But Claude Code won't launch without paid access. If you don't have an account yet, first find out which email address to use to register: you can't change it later.

How do you check that everything installed correctly?#

There are three commands worth knowing from the start.

The first checks the version:

claude --version

A working installation responds with a number in the format X.Y.Z (Claude Code). The specific digits change from version to version; the important thing is that the command returns a version number at all. A “command not found” response means something else—I'll cover it in the next section.

The second is diagnostics, if something goes wrong.

The third is /status inside a session: it shows how you signed in, which email or organization is linked, and whether your access has expired. And claude update applies the latest update immediately instead of waiting for the background update.

Why can't the claude command be found after installation?#

This is the most common complaint from beginners, and it sounds scarier than it is. The terminal doesn't search the entire computer for commands; it only checks the folders in PATH. Claude Code puts itself at ~/.local/bin/claude (on Mac and Linux) or %USERPROFILE%\\.local\\bin\\claude.exe (on Windows), and if that folder isn't in PATH, the terminal simply can't see the program.

On Mac and Linux, add the folder to the list. Open your shell's settings file (~/.zshrc for Zsh, ~/.bashrc for Bash) and add this line at the end:

export PATH="$HOME/.local/bin:$PATH"

Then close and reopen the terminal (or run source ~/.zshrc). After that, claude will be found.

Common installation and first-launch errors#

Let's go through them in order; all are from Anthropic's official troubleshooting guide.

1. Old Node when installing through npm. The @anthropic-ai/claude-code package requires Node.js 22 or later. On Windows, npm may fail with a message about scripts being disabled—PowerShell's policy is blocking them from running. In WSL, a common error is exec: node: not found: WSL is using Windows Node instead of Linux Node. If npm is giving you trouble, switch to the native installer; it doesn't need Node.
2. Wrong installer for the shell. 'irm' is not recognized means you pasted a PowerShell command into Command Prompt. The token '&&' is not a valid statement separator means you did the opposite and ran a CMD command in PowerShell. A parameter cannot be found that matches parameter name 'fsSL' means you ran the Mac/Linux command in PowerShell. Check the start of the prompt: PS C:\ is PowerShell, and C:\ is CMD.
3. An old API key overrides the subscription. If your subscription is active but an old key is still set in your environment, you'll see API Error: 400 ... This organization has been disabled. The key takes precedence over subscription sign-in. Remove it with unset ANTHROPIC_API_KEY (in PowerShell, use Remove-Item Env:ANTHROPIC_API_KEY), check ~/.zshrc and ~/.bashrc, then type /status.
4. Insufficient memory on a weak server. Installation stops with the word Killed and code 137—that means there isn't enough RAM. You need about 512 MB free; adding swap fixes it.
5. The installer returned something else. The error syntax error near unexpected token '<' or a reference to <!DOCTYPE html> means the installation URL returned a web page instead of the installation script—usually because of your region, a corporate firewall, or a temporary outage. Try another method (Homebrew, WinGet) or try again in a few minutes.

For any of these errors, the first step is the same: run claude doctor. It will show what's broken and suggest where to look.

There's a diagnostic wrinkle here: a bare 403 during installation can also come from a corporate firewall or proxy. So if you're on an office network, first check whether it's blocking access.

What to do after installation: first steps#

Installation is just the beginning. The real work starts next, and the first thing people run into is memory. A language model doesn't remember you between conversations: in a new chat, it knows only what you've put in its context.

So the first real step after installation is to create a CLAUDE.md file in your project's root. It's a text file with project rules and context that Claude Code reads every time it launches. Describe once what you do, how the project is organized, and what matters—and you won't have to repeat it in every conversation.

I walk through how to put that file together and customize Claude in a separate guide: how to customize Claude. I also put together a general map of vibe-coding tools for 2026.

What else do people ask?#