How to Install Claude Code and Start Working in 5 Minutes

Claude Code is Anthropic's coding agent for the terminal. It can read project files, propose changes, edit code, and run commands. To start safely, install the current CLI, open a small test project, sign in, and give the agent a read-only task first. For the first edit, keep manual confirmation enabled and always inspect git diff.

Quick summary

  • The current recommended installation path is Native Install, not the old global npm package.
  • Start it from a project folder with claude.
  • For the first session, keep Manual permissions, or begin in Plan mode.
  • First ask Claude to explain the project without changing anything.
  • Limit the first edit to one file, then inspect git diff.
  • Use Ctrl+D to exit; Ctrl+C stops the current operation.

What Claude Code is

Claude Code is not just a browser chat. The agent receives access to your working folder and can read code, edit files, execute commands, and work with development tools. It is available through the terminal, IDE integrations, desktop surfaces, and browser-based workflows.

This guide uses the Terminal CLI because it makes the workflow easiest to understand: open a project, describe a task, review the requested action, approve it, and verify the result.

What you need before installation

Prepare the basics:

  • a terminal: Terminal or iTerm on macOS, a Linux shell, PowerShell/CMD, or WSL on Windows;
  • a small project that is safe to modify;
  • Git, ideally with a clean git status;
  • a Claude Pro, Max, Team, or Enterprise account, an Anthropic Console account, or a supported external provider.

Claude Desktop, the Terminal CLI, and VS Code support third-party providers in different ways. If you connect Claude Code through BetterToken from Russia, use the Russian guide for API Key and Anthropic-compatible Base URL in the BetterToken documentation. Installing the application or CLI and configuring the provider are separate steps: first install the client, then add access parameters.

How to install Claude Code

Anthropic recommends Native Install. The current quickstart uses these commands.

macOS, Linux, and WSL

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

Windows PowerShell

Use the current Windows command from the official Claude Code quickstart. If your organization manages installers centrally, follow the internal policy instead of running random install commands copied from social media.

After installation, open a new terminal and check:

claude --version

If the command is not found, restart the terminal and check whether the installer added Claude Code to your shell path.

First safe launch

Create or open a small project, then run:

cd path/to/test-project
git status
claude

For the first message, do not ask it to refactor anything. Ask for orientation:

Read this project and explain its structure. Do not edit files or run commands yet.

This gives you a low-risk first interaction. You can see whether Claude understands the repository before allowing edits.

First tiny task

Pick one harmless change. For example:

Update the README with a short "Local development" section. Touch only README.md. Ask before running commands.

When Claude asks for permission, read the action carefully. If it wants to edit unrelated files or run a broad command, stop and narrow the task. After it finishes, run:

git diff

Review the exact change. If the result is wrong, ask Claude to revise it while keeping the same file boundary.

Permission modes for beginners

Start with Manual mode. It may feel slower, but it teaches you how Claude Code thinks and what kinds of actions it requests. Plan mode is also a good first-session habit when the task touches more than one file.

Avoid auto-approving broad actions until you know the project, the agent behavior, and the command safety profile. The goal of the first day is not maximum speed. It is learning how to supervise the tool.

Connecting through an API provider

If your workflow uses a provider or gateway, you usually need three ideas:

  • API Key: the token that authenticates you;
  • Base URL: the Anthropic-compatible endpoint the client calls;
  • Model: the model name or alias supported by that provider.

For BetterToken, use the provider setup page or documentation so the exact Base URL and key format match the current platform. Do not paste tokens into public screenshots, project files, or shared shell history.

Common first-run problems

If claude is not found, the shell path probably has not refreshed. Open a new terminal and check the install location.

If login fails, confirm whether you are using a Claude account login, Anthropic Console credentials, or a third-party provider flow. Those are different paths.

If a provider request fails, check the API Key, Base URL, model name, and whether your network or proxy changes the request.

If Claude edits too much, reduce the task size. Give file boundaries, ask it to plan first, and keep manual approval on.

A good first workflow

The safe beginner workflow is simple:

Open project -> ask for explanation -> ask for a plan -> approve one small edit -> inspect git diff -> commit or revert.

Once this feels natural, you can move to larger tasks: tests, bug fixes, refactors, documentation, and controlled automation. Claude Code becomes useful fastest when you treat it like a junior teammate with strong reading and writing skills, not like a magic button that should be left alone in a production repository.