How to Install and Set Up Claude Code: Step-by-Step Guide

Two things determine your setup path before you touch an installer:

• How you will pay. Claude Code is available through Claude Pro and Max subscription plans, or via the API on a pay-as-you-go basis. Subscriptions suit individuals and small teams who want predictable billing; API access suits organisations that want central billing, usage controls, and programmatic key management. Check Anthropic’s pricing page for current terms.
• Your environment. For the CLI you need a supported terminal on macOS, Windows, or Linux. Node.js is required for the npm-based install path. Git should be installed and configured, since Claude Code leans on it heavily for branching, commits, and safe experimentation.

If you are evaluating for a team, decide up front whether engineers authenticate with individual subscriptions or organisation-managed API keys — retrofitting this later is the most common source of billing confusion.

The CLI is the original and most flexible way to run Claude Code — terminal-native, scriptable, and usable over SSH on remote development machines.

• Install via the official installer script or npm package from Anthropic. On Windows, it runs natively or under WSL; on macOS and Linux, any standard shell works.
• Verify by running claude in a terminal — you should see the interactive session start.
• Authenticate on first launch: the CLI walks you through logging in with your Claude account (Pro/Max) or configuring an API key.

A note for locked-down machines

If your organisation restricts installers or outbound traffic, involve IT early: Claude Code needs network access to Anthropic’s API endpoints. Enterprise deployments can also route through supported cloud providers — worth knowing before your pilot stalls on a firewall ticket.

Not everyone on an engineering team wants a terminal workflow, and Claude Code no longer requires one.

• Desktop app (Mac and Windows). Download from Anthropic, sign in, and point it at a local repository. You get the same agent with a visual interface: session management, side-by-side diff review, and parallel sessions across projects. It is often the best on-ramp for engineering managers and less terminal-centric developers.
• Web (claude.ai/code). Run Claude Code from a browser against connected repositories. This is ideal for kicking off well-scoped tasks — a bug fix, a dependency bump, a first draft of a feature — from anywhere, then reviewing the resulting changes as a pull request.

The important point for team rollouts: these surfaces share the same project memory and configuration as the CLI, so a mixed team does not fragment into incompatible setups.

For engineers who want the agent inside their editor, Claude Code ships extensions for VS Code and JetBrains IDEs (IntelliJ, PyCharm, WebStorm, and family).

• Install from the VS Code Marketplace or JetBrains Plugin Marketplace, then sign in with the same account you use elsewhere.
• The extension runs full agentic sessions in a panel beside your code, with inline diffs you can accept or reject file by file.
• Context flows naturally: the file you have open and your current selection are visible to the agent, which shortens prompts considerably.

CLI or extension — which should a team standardise on?

Neither, deliberately. Let engineers choose their surface and standardise the layer underneath: shared CLAUDE.md files, permission settings, and MCP configuration. That is the layer that determines output quality, not the window the session runs in.

The install is the easy half. These three steps determine whether Claude Code behaves like a disciplined colleague or a loose cannon:

• Authentication and model choice. Confirm you are on the intended plan and select a default model — the flagship Claude Fable 5 for complex work, Sonnet or Haiku-class models for fast routine tasks. See our Claude model guide for how to choose.
• Permissions. Claude Code asks before editing files or running commands. Tune the allowlist: pre-approve safe, frequent operations (running tests, formatters) and keep approvals on anything destructive or network-facing.
• CLAUDE.md. Create this file at your repo root. Document build and test commands, architecture conventions, style rules, and known pitfalls. Every session reads it automatically — it is how one engineer’s hard-won context becomes everyone’s default.