Getting Started with Claude CLI

Most people meet AI through a chat window: you type a question, you get an answer, and then you copy the answer somewhere useful. Claude Code (the "Claude CLI") flips that around. You run one command in a project folder and Claude can read your files, search your code, run commands, and edit things, all while you watch and approve. It is less like asking a librarian a question and more like handing a capable colleague the keys to your workshop and saying "let's fix this together."

This week we'll get it installed on Windows and macOS, learn the handful of commands you will use every day, and talk honestly about what it is, and what it isn't.

What You Need First

• An account that includes Claude Code. That means a Claude Pro, Max, Team, or Enterprise plan, or an Anthropic Console account. The free claude.ai plan does not include it.
• A reasonably recent computer: macOS 13+, Windows 10 (1809) or newer, 4 GB of RAM or more, and an internet connection.
• A terminal. On a Mac that's the Terminal app. On Windows it's PowerShell (search the Start menu for it).

Rough cost of each plan (US dollars; see claude.com/pricing for current numbers):

Claude Code shares your plan's usage limits with chat, so a heavy coding day can hit them. That's when people step up to Max or switch to pay-as-you-go.

You do not need Node.js, and you don't need administrator rights on Windows.

Install on macOS

Open Terminal and paste:

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

Prefer Homebrew? That works too:

brew install --cask claude-code

The native installer updates itself in the background. Homebrew installs do not, so run brew upgrade claude-code now and then.

Install on Windows

Open PowerShell and paste:

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

Or, if you like WinGet:

winget install Anthropic.ClaudeCode

(WinGet installs don't auto-update either: run winget upgrade Anthropic.ClaudeCode occasionally.)

Using the old-style Command Prompt instead? Use this one-liner there:

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

Tip: If PowerShell complains that && isn't a valid statement separator, you pasted the CMD command into PowerShell. If it says irm isn't recognized, you pasted the PowerShell command into CMD. Look at your prompt: PS C:\> means PowerShell.

Check That It Worked

Close the terminal, open a fresh one, and run:

claude --version

You should see a version number. If your shell says claude can't be found, the install folder isn't on your PATH yet, and the official troubleshooting guide walks through the fix. For a deeper health check, run claude doctor.

Your First Session

Go to a folder you want to work in, then start Claude:

cd my-project
claude

The first time, it opens a browser window so you can sign in. After that you land in an interactive prompt. Just type in plain English:

What does this project do? Give me a quick tour of the main files.

Claude will read the files it needs and answer. Try something small next, like "add a --verbose flag to the script". Before it changes a file or runs a command, it asks permission. You see exactly what it wants to do and you say yes or no. That approval step is the heart of the whole tool.

The Commands You'll Actually Use

Type / at the start of the prompt and Claude shows every command available. These are the everyday ones:

Three Prefixes and Four Keys Worth Memorizing

• / at the start runs a command, as above.
• ! at the start runs a shell command directly, with no AI involved, and adds the output to the conversation. !git status shows Claude what you're looking at.
• @ mentions a file. Type @ and start a filename, then press Tab to autocomplete. Claude reads it.

And the keys:

• Esc stops Claude mid-thought so you can redirect. It keeps the work done so far.
• Esc twice on an empty prompt opens the rewind menu.
• Shift+Tab cycles permission modes: normal (ask every time), accept edits, and plan mode, where Claude explores and proposes a plan but changes nothing until you approve it. (On Windows, if Shift+Tab doesn't register, try Alt+M.)
• Ctrl+C interrupts a running operation; press it again at an empty prompt to quit.

Make It Smarter About Your Project: CLAUDE.md

Run /init once and Claude writes a CLAUDE.md file in your project. Edit it freely. It's where you put the things you'd otherwise repeat every day: "we use tabs, not spaces," "never edit the generated files," "tests run with npm test." Claude reads it at the start of every session, so it behaves like a teammate who remembers your house rules. This one file is the biggest quality jump for the least effort.

Windows Bonus: Give It Linux Tools

Claude Code works fine on plain Windows. Without anything extra it runs commands through PowerShell. But it was born in a Linux-and-Mac world, and a lot of its instincts are shell tools like grep, sed, awk, find, head, and tail. Give it those on Windows and things go smoother, especially for searching and text-munging.

There are two good ways to do that:

1. Install Git for Windows (free, optional but recommended). It includes Git Bash, and when it's installed Claude Code automatically uses Bash for its shell tool. It also gives you git itself, which you want anyway. If Claude can't find it, point to it in your settings.json:

   {
    "env": {
      "CLAUDE_CODE_GIT_BASH_PATH": "C:\\Program Files\\Git\\bin\\bash.exe"
    }
   }

2. Use WSL (Windows Subsystem for Linux). Run wsl --install in an admin PowerShell, then install and run claude inside the Linux terminal. You get a full Linux environment and Claude's sandboxing support.

I also keep a folder of GNU tools compiled for Windows (grep, sed, awk, curl, tar, less and friends) on my PATH in D:\bin. That's a convenient trick for running those tools from a plain PowerShell or CMD window, but honestly, Git for Windows already covers the same ground and is the easier route. Reach for a standalone tool folder only if you want those commands available everywhere rather than just inside Claude.

Claude Is a Tool, Not a Threat

Let's address the elephant in the terminal. Every few months a headline announces that AI is coming for programmers, writers, or whoever's turn it is. I use Claude every day, and here is my honest take: it is a power tool.

A table saw makes a carpenter faster and more ambitious. It does not decide what to build, check whether the house should have a second floor, or take responsibility when something goes wrong. Claude is the same:

• You stay in charge. Nothing happens without your approval unless you pre-approve it. You can read every command before it runs.
• It makes mistakes. It can be confidently wrong, so you read the diffs, run the tests, and use your judgment. The more you know, the better you steer it.
• It's only as good as your direction. Clear goals and good context get great results; vague requests get vague results.
• It removes the tedious part. The boilerplate, the "what was that flag again," the first draft of the repetitive thing. That leaves more of your day for design and decisions.

The people who benefit most aren't the ones who hand over their brain. They're the ones who learned what to ask for.

How I Use It

A few real examples from my own work, so you can picture it:

• Working in a large, unfamiliar codebase. WaSQL is a big PHP framework with years of history. Instead of grepping for twenty minutes, I ask "where does the chart tag processor build its color palette?" and Claude finds it, reads it, and explains it.
• Tracking down a real bug. Just this week a page on this very site threw a "modulo by zero" error. Claude traced it through the framework to a loop that was quietly wiping a shared color list, explained why, proposed a one-line-per-place rename, and I reviewed and shipped the fix.
• Building this website. The very articles you read here are managed by pages Claude helps me build and maintain, including the admin reports I use to watch the traffic trend.
• Building a whole product. ClickBooks, a free picture-book site that teaches languages, plus its Roku channel and a Google TV wrapper. If you have a Roku, you can check it out at https://my.roku.com/add/clickbooks - it is already live. I'd never built a Roku channel before. Claude helped with the code, the narration pipeline, the translations, and the tedious packaging steps.
• Chores. Renaming things across dozens of files, writing a script to resize a pile of images, converting a document format, drafting a migration. The jobs I'd otherwise put off.
• Writing and checking. It even helps me research and fact-check articles like this one, though I make sure every command in here is one you can verify yourself in the official docs.

Good Habits for Day One

1. Start in a git repository. Then every change Claude makes is easy to review or undo.
2. Be specific. "Fix the login bug where an empty password gets accepted" beats "fix login."
3. Use plan mode for anything big. Let it propose before it acts.
4. Run /clear between unrelated tasks so old context doesn't muddy new work.
5. Read what it does. Ten seconds of reading a diff is the best insurance there is.
6. Run /init and keep CLAUDE.md current.

Your Challenge This Week

Install it, open a project you know well, and ask this: "Give me a tour of this project, and tell me one thing you'd improve." Then argue with the answer. Disagreeing with Claude is half the fun, and it's how you learn where it's strong and where you still need to be the expert.

P.S. — Claude Code also runs inside VS Code and JetBrains editors, in a desktop app, and on the web, all from the same account. Once the terminal version feels comfortable, the rest is a short hop. The full documentation is excellent.