Getting Started with Claude Code: From Zero to First Real Task

What You Are Getting Into

Claude Code is a CLI tool that drives a Claude session against your actual codebase. It can read files, run commands, use git, and write code — all while showing you what it plans to do before it does it. You approve or redirect each step.

The gap between Claude Code and a chatbot is significant. A chatbot answers questions about hypothetical code. Claude Code works on the files sitting in your project directory right now. That changes what it can do.

Install It

On macOS, the fastest path is Homebrew:

brew install anthropic/claude-code/claude-code

On Windows, download the installer from the Claude Code website and run it. On Linux, download the tarball and move the binary into your PATH:

tar -xzf claude-code-linux.tar.gz
sudo mv claude /usr/local/bin/

Verify the install:

claude --version

If you see a version number, you are ready.

First Authentication

Run claude in any terminal. It opens a browser window for Anthropic login if this is your first time. Log in with your Anthropic account or create one. After that, the CLI uses a cached token — you will not need to authenticate again unless you explicitly log out.

For CI or server environments, set ANTHROPIC_API_KEY instead of using browser login. The CLI checks for this environment variable before attempting the browser flow.

Start Your First Session

Navigate to a project you understand well. Not a toy project — something real where you know what the code does and what a good outcome looks like. A bug you have been putting off, a refactor you have been meaning to do, tests you have been avoiding.

cd ~/projects/your-project
claude

Give it a clear task. The clearer you are, the better the output:

Write tests for the auth module. Focus on token validation and refresh token flow. Do not modify the source files.

Notice the constraints: what to focus on, what not to change. Constraints tell Claude Code what the boundaries of the task are.

Watch and Learn

Claude Code shows you its plan before executing each step. Read what it intends to do. Approve or redirect — it does not touch your files until you say go. This is the feedback loop that teaches you how Claude Code thinks about your codebase.

If it starts in the wrong place, say so. If the approach is wrong, interrupt and redirect. You are teaching it your project context and your preferences. The more you interact, the better it gets.

Review What It Produces

When the session ends, review the changes:

git diff

Look at what Claude Code wrote. Ask yourself: is this what I would have written? If not, what is different? The answers to those questions teach you when to trust Claude Code's judgment and when to apply tighter constraints.

This first session teaches you more about how Claude Code works than any documentation. You will have a feel for what to be specific about, where it needs boundaries, and how it approaches code.

What Comes Next

After your first session, you know enough to be dangerous. The next thing to learn is the configuration system — how to set a default model, configure hooks, and adjust behavior per-project. But you do not need that on day one.

Day one is: install Claude Code, run it against a real task, see what it produces. Everything else builds from there.