Installing Claude Code

Claude Code runs on macOS, Linux, and Windows via WSL. Native Windows (outside WSL) is not supported. There are three installation routes -- use whichever fits your workflow.

Installing Claude Code is not the same as installing an app from the App Store or downloading a program. Claude Code is a command-line tool -- software that runs in a terminal window (a text-based interface where you type instructions rather than clicking buttons). If you have never used a terminal before, that sounds intimidating. In practice, you type one command to install it and one command (claude) to start it. That is the extent of the terminal work for most users.

Option 1 -- Native installer (recommended)

The native installer handles updates automatically and is the simplest path on macOS and Linux.

# macOS / Linux
curl -fsSL https://claude.ai/install.sh | sh

Option 2 -- Package managers

# macOS with Homebrew
brew install claude-code

# Windows (WSL) with WinGet
winget install Anthropic.ClaudeCode

Option 3 -- npm (global package)

npm install -g @anthropic-ai/claude-code

npm (Node Package Manager) is a system for distributing software built in JavaScript. If you use Option 3, you will first need Node.js installed -- Node.js is the underlying software that runs JavaScript programs on your computer, outside of a browser. You can check if you already have it by typing node --version in a terminal. If nothing comes back, install Node.js from nodejs.org before running the npm command above. Options 1 and 2 handle this automatically, which is why they are simpler for most people.

Requirements

An Anthropic account on a Pro, Max, Teams, or Enterprise plan
macOS 13+, Ubuntu 20.04+, Debian 10+, or Windows with WSL installed
npm route only: Node.js 18 or later -- check with node --version
Windows WSL route: Git for Windows must be installed before setting up WSL

The native installer (Option 1) and package managers (Option 2) handle all dependencies automatically -- no separate Node.js install required.

First run and authentication

Run claude in any directory. On first launch you will be prompted to sign in with your Anthropic account via browser OAuth -- a secure sign-in process where a browser window opens, you confirm your identity, and the connection is established automatically. Once authenticated, the session token is stored locally -- you will not need to sign in again unless you explicitly log out.

cd ~/my-project
claude

What to expect on first run

The first time you run Claude Code, a few things happen in sequence:

A browser window opens asking you to authorise access to your Anthropic account -- sign in as you would on claude.ai
The terminal shows a confirmation that authentication succeeded
Claude Code presents a short welcome prompt and asks what you would like to work on
You are now in an active session -- type a question or instruction to begin

The first session will feel similar to using Claude in a browser, except Claude can now read files in the folder you launched it from. Try asking: "What files are in this directory?" Claude will list them. That is the difference -- it is operating directly inside your computer's file system, not in a sandboxed browser tab.

Permission model -- how Claude asks before acting

Claude Code operates on a tiered permission system. Think of it like a new employee who checks with you before taking certain actions rather than going ahead independently. Here is how it works in practice:

Read operations (reading files, searching for content) -- allowed without prompting. Claude can look at your files without asking each time.
File edits and writes -- Claude proposes a change (called a "diff" -- a side-by-side comparison showing exactly what would be added, removed, or changed in a file) and waits for your approval before anything is written to disk. You see precisely what will change before it happens.
Bash and shell commands (running programs, terminal operations) -- Claude shows you the exact command before running it. You approve or deny. Nothing runs without you seeing it first.
MCP tool calls (connections to external services like your calendar or email) -- require explicit approval per tool unless you have pre-approved them in settings.

Permissions in plain English

Claude asking for permission before editing files or running commands is not a bug or an inconvenience -- it is a core safety feature. Think of it as Claude behaving like a careful employee who confirms with you before sending an email on your behalf, rather than one who acts unilaterally and hopes you approve later. You can adjust how often Claude prompts you as you build confidence in its behaviour.

You can switch permission modes during a session. The default "Ask" mode prompts for everything sensitive. "Auto-accept edits" skips per-file approval for the session. These are controlled via the mode selector in the panel or the /permissions command.

macOS -- no extra permission prompts

Claude Code runs as a standard CLI tool on macOS. No Keychain prompts, no file system access dialogs, no special OS-level authorisations. Git is pre-installed on macOS; if it is not, macOS will offer to install Xcode Command Line Tools the first time a git command runs -- this is a standard macOS prompt, not a Claude Code requirement.

VS Code extension

Install the official Claude Code extension from the VS Code marketplace. Once installed, Claude Code runs in a dedicated side panel -- files open in the editor are directly visible to Claude without needing to navigate the file system. See the full guide: Using Claude Code in VS Code.

JetBrains plugin

Available for IntelliJ IDEA, PyCharm, WebStorm, and other JetBrains IDEs. Install via Settings → Plugins → Marketplace, search "Claude Code". The plugin provides the same embedded panel experience as VS Code.

Keeping it updated

# npm install
npm update -g @anthropic-ai/claude-code

# Homebrew
brew upgrade claude-code

Claude Code updates frequently. The native installer and Homebrew manage updates automatically. When running via npm, update periodically or watch for in-session prompts.

Windows note

WSL is the only supported Windows path. Git Bash is not supported. Install WSL first (wsl --install in PowerShell as administrator), then install Claude Code inside the WSL terminal.

Troubleshooting common install issues

Problem	Likely cause	Fix
`command not found: claude`	Installation did not add Claude to your system path	Close and reopen the terminal, then try again. If still missing, restart your machine.
`npm WARN` or permission errors on npm install	npm global permissions misconfigured	Use Option 1 (native installer) or Option 2 (Homebrew) instead to avoid npm permission issues entirely.
Browser opens but authentication hangs	Browser popup blocked or OAuth callback port blocked by firewall	Allow popups for claude.ai in your browser settings. On corporate networks, check with IT about outbound port restrictions.
macOS: "app cannot be opened because it is from an unidentified developer"	macOS Gatekeeper blocking unsigned binary	Right-click the installer and choose Open, then confirm. This only needs to be done once.
Node.js version error during npm install	Node.js version below 18	Update Node.js at nodejs.org, then reinstall. Check your version with `node --version`.

What comes after installation

A fresh Claude Code install is functional but unconfigured. Claude does not know who you are, what conventions to follow, or how to behave in your workspace. The next step is setting up your workspace: a CLAUDE.md file with your context, rules files for behavioural constraints, and optionally MCP servers for external tool access. See Your First Claude Code Workspace for how to build this from scratch.

If you would rather start with a complete, pre-built workspace configuration, AI Business OS provides a ready-made setup -- CLAUDE.md, rules library, skills framework, MCP server configuration, and session management -- designed to be cloned and adapted rather than built from nothing. It is the fastest path from a bare Claude Code install to a fully functioning personal assistant environment.