Up To Speed With AI homeFind my AI

Independent · No affiliate links

Guides › New to AI

Claude Code for beginners: how to work on a real project

Updated · Level: Beginner · Facts from Anthropic's own docs. Tips and example prompts are ours.

You don't have to be a developer to use Claude Code on a real project. This page starts with how to install it and open it, including Windows PowerShell if that's how you need to run it, then covers how to ask, how to get a plan before anything changes, how to undo a mistake, and how to stay safe. Which plans include it, and a shorter setup recap: Claude Code explained.

Get it installed and open

You only do the install once. After that, you just open the app or type claude in a terminal. Anthropic's own terminal guide for new users2 is the source for the PowerShell and Terminal steps below. 2,6,1

What you need

Choose the desktop app or the terminal

There are two common ways to use Claude Code on your computer:

You can install both. They can work on the same project folder. 1

Option A: the desktop app, with no terminal

  1. Go to the Claude download page9 and download the installer for Mac or Windows. Run the file to install the app. 9,10
  2. Open Claude from your Applications folder on a Mac, or from the Start menu on Windows, and sign in with your paid account. 1,10
  3. Click the Code tab at the top of the sidebar. If you only see icons, hover to see the names. If it asks you to upgrade, you need a paid plan first. 1
  4. Under the message box, choose Local, then Open folder… and pick your project folder. Start with a small folder you know well, or a copy of it. 1

What you'll see: a chat box, plus controls for the folder and how much Claude can do without asking. In Manual mode it shows you each change and waits for you to approve it. 1

Option B, on Windows: install and open it in PowerShell

You need Windows 10 version 1809 or later. You do not need to run PowerShell as Administrator. 6,2

Anthropic recommends installing Git for Windows as well, so Claude Code can use its Bash tool. It's optional: without it, Claude Code uses PowerShell instead, and you don't have to learn Git. If you don't already have it, download the installer from git-scm.com/downloads/win, run it, and click Next on each screen to keep the defaults. When you see "Adjusting your PATH environment," leave the recommended option selected. If you already have Git, you can skip this. 6,2

  1. Open PowerShell. PowerShell is Windows' built-in window for typing commands; it comes with Windows. Open the Start menu, type PowerShell, and open Windows PowerShell or Terminal — not the one that says (x86). You can also hold the Windows key and press X, then choose Windows PowerShell (or Terminal). A window with a blinking cursor appears. 2
  2. Check you're in PowerShell, not Command Prompt. The line should start with PS C:UsersYourName>. If you only see C:UsersYourName> without PS, you're in CMD. Close it and open PowerShell, or use the CMD install command in the note below. 2,6
  3. Paste the official install command with Ctrl+V or a right-click, then press Enter. The command shows no progress while it downloads, so leave the window open and wait. When it finishes, you should see "Claude Code successfully installed!" 2,6

Windows PowerShell — copy this line, paste it, and press Enter

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

Only if you're in Command Prompt (CMD), not PowerShell

curl -fsSL https://claude.ai/install.cmd -o install.cmd && install.cmd && del install.cmd
  1. Open a new PowerShell window (close the old one, then open PowerShell again) and type claude --version, then press Enter. A working install prints a version number, such as 2.1.211 (Claude Code). 6

Check that it installed

claude --version

Option B, on a Mac: install and open it in Terminal

You need macOS 13 or later. 2,6

  1. Open Terminal. Press Cmd+Space to open Spotlight, type Terminal, and press Enter. A window with a blinking cursor appears. That's where you type commands. 2
  2. Paste the official install command with Cmd+V, then press Enter. It shows no progress while it downloads; leave the window open and wait. When it finishes, you should see "Claude Code successfully installed!" 2

Mac, Linux or WSL — copy this line, paste it, and press Enter

curl -fsSL https://claude.ai/install.sh | bash
  1. Open a new Terminal window and type claude --version, then press Enter. A working install prints a version number followed by (Claude Code). 6

On most Linux systems you can open a terminal with Ctrl+Alt+T, or search for Terminal, then use the same install command as on a Mac. 2

The first time you open it: sign in

In the terminal, go to your project folder (next section) and type claude, then press Enter. The first time, Claude Code opens a browser window so you can log in with your claude.ai account. If the browser doesn't open, press c to copy the login URL and paste it into your browser. If the browser shows a login code instead of sending you back, paste that code into the terminal where it says Paste code here if prompted. When login finishes, the terminal shows Login successful and asks you to press Enter to continue. 8,11,2

After that, your sign-in is stored, so you shouldn't have to log in every time. To sign out later, type /logout at the Claude Code prompt. 11,8

In the desktop app, you sign in when you first launch Claude, before you open the Code tab. 1

What you'll see in the terminal: after you're signed in, a text prompt appears, with the version, the current model and your working folder shown above it. You type a message and press Enter to send it. You can't click around in the terminal; use the arrow keys. Type /help to see commands. 11,2

Get into your project folder

In the desktop app, use Open folder… as described above. In the terminal, you first move into the folder, then start Claude Code. Officially that looks like this11

Go to your project folder, then start Claude Code

cd /path/to/your/project
claude

cd means "change directory": it points the terminal at a folder on your computer. Replace /path/to/your/project with the real path of your folder. If the path has spaces, put quotes around it, for example cd "C:\Users\YourName\Documents\my project". 11

To copy that path:

Then type claude and press Enter. Work in a copy of the folder if you're just getting started, so a change is easy to undo.

How to reopen it next time

How to quit

In the terminal, type exit or /exit, or press Ctrl+D twice on an empty prompt. That leaves Claude Code and takes you back to the ordinary PowerShell or Terminal prompt. 2,11

In the desktop app, close Claude the way you close any other app.

Keeping it up to date

If you installed it with the commands above (Anthropic calls this a native install), Claude Code updates itself in the background. Updates take effect the next time you start it. To update immediately, run claude update in a terminal (not inside a Claude Code session). 6

Update Claude Code now

claude update

If something goes wrong

Check your install

claude doctor

Your first 5 minutes

  1. Confirm you're on a paid Claude plan and can sign in at claude.ai. 7
  2. Install the desktop app from claude.com/download, or run the official install command in a new PowerShell or Terminal window, then check it with claude --version. 9,6
  3. Sign in when the browser window opens (or when the desktop app asks). 8
  4. Open your project folder: Open folder… in the desktop app, or cd to the folder and type claude in the terminal. 1,11
  5. Ask it to look around and explain the folder in plain English, and to change nothing yet. That's the first prompt later on this page.

Before you start work

Once it's installed and open, a few habits will keep you safer. Claude Code shares your plan's limits with your chats. 7,6 If you want the shorter setup recap, it's in Claude Code explained.

How to describe what you want

Talk to it the way you'd brief a careful helper who has never seen the project. Include:

  1. What you want done, in one or two sentences.
  2. What's already there, if you know (a website, a form, a folder of files).
  3. What not to touch (for example: "don't change the design, only the wording").
  4. How you want it to talk to you: "Explain each step in plain words, and ask before you change anything."
Prompt: describe a job
Look around this folder and explain in plain English what this project is and how it fits together.
Don't change anything yet. Then tell me the smallest useful change I could ask for first, and what I'd need to check afterwards.
Prompt: ask it to explain the steps
I want to [describe the job]. Before you change any files, explain the steps in plain words, which files you'd touch, and anything that could go wrong. Wait for me to say go ahead.

Ask it to plan before it changes anything

Plan mode tells Claude to look around and propose a plan without editing your files. You can turn it on by prefixing a message with /plan, or with the mode control (in the terminal, Shift+Tab). When the plan is ready, you can approve it, stay in planning and ask for changes, or review each edit yourself. 3

A plain-English request works too, and is a good habit even if you forget the command:

Prompt: plan first, don't change yet
Don't change any files yet. Make a short plan for [the job], in plain words: the steps, the files you'd touch, and how we'd check it worked. Stop and wait for my go-ahead.

Permission prompts: read them before you say yes

A permission mode sets what Claude can do without asking you first. In Manual mode it asks before most actions that edit files, run commands or use the network. Plan mode explores and writes a plan without editing your source. Accept edits lets file edits through and still asks about many commands. Auto mode lets a second model review actions instead of you. You can change the mode during a session. 3

If you want to see every step, start in Manual or Plan. Anthropic says you're responsible for reviewing proposed code and commands for safety before you approve them. 4

When a prompt appears, read what it's about to do. If you don't understand it, type "Explain this in plain words, and what happens if I say no." You can refuse anything that looks like it's deleting files, sending data, or installing software you didn't ask for.

Checkpoints, undo and git

Claude Code automatically tracks the file edits it makes in a session. To undo them, run /rewind or its alias /undo. You can restore the files, the conversation, or both, back to an earlier prompt. 5

That is a session safety net, not a lasting history. A file that was only changed by a shell command, or by something outside this session, may not be undone that way. Anthropic says checkpoints are not a replacement for version control such as git. 5

If the project already uses git (you'll often see a hidden .git folder, or a site like GitHub), ask Claude to save a snapshot — a "commit" — before a big change, and again after you've checked the result. If it doesn't use git yet, working in a copy of the folder is the simplest backup. You can also ask: "Show me how to save a git snapshot of the current files, in plain words, and wait for me to say go ahead."

Reading its output and errors

You don't have to understand every line of code. You do need to follow the story:

When something fails, paste the error back and ask it to explain. Don't keep saying "try again" without understanding the failure.

Prompt: explain an error
Here's the exact error: [paste]. Explain in plain words what it means, what you were trying to do, and the safest next step. Don't change anything until I say so.

Keep notes in CLAUDE.md

CLAUDE.md is a plain-text notes file in your project that Claude Code reads at the start of every session: your rules, preferences and how the project works. 12 You can type it yourself, or run /init to have Claude draft a starter file from what's already in the folder. If the file already exists, /init suggests improvements rather than overwriting it. 12

Write down the things you keep repeating: "Explain in plain words", "Plan first and wait", "Don't deploy without my go-ahead", and any project-specific rules. Anthropic suggests keeping it short and specific; the more concise the instructions, the more consistently they're followed. 12,13

Prompt: start a notes file
Create or update a CLAUDE.md in this folder with short, plain-English notes for future sessions: what this project is, how I like you to work (plan first, explain in plain words, ask before changing anything), and anything I shouldn't want you to touch. Show me the draft and wait for my go-ahead before saving it.

When you're stuck

Prompt: when you're stuck
Stop and don't change any more files. In plain words: what we've done, what still isn't working, what you're unsure about, and three options for what to try next, safest first. Wait for me to pick.

Safe habits

Prompt: review before anything goes live
Don't publish, deploy or push anything. List every file you changed and, in plain words, what each change does. Tell me how I can check the result myself, and anything that still looks risky.

When the work is bigger than one session: Projects

In Claude Code, a project is one ongoing conversation where Claude coordinates a stream of related tasks. It starts a parallel thread for each piece of work, answers questions in the project chat, and shows you which threads need you. Most threads run in the cloud and keep going after you close your laptop; if a task needs something only your computer has, you can ask Claude to run that thread on your computer through Remote Control. 14,15

Projects are in public beta on Pro, Max and Ultra, rolling out gradually, and aren't offered on Team or Enterprise yet. How to start one, and a copyable first prompt: Projects in the Claude Code guide.

Want to leave Claude Code running on your computer and steer it from your phone or a browser? That's Remote Control, and it needs the computer to stay on. Control Claude Code from your phone or the web.

Want the main model to do the everyday work and call a stronger model only for hard decisions? That's the Advisor in Claude Code.

Common questions

5 quick questions, answered

Do I have to use PowerShell or a terminal?

No. The Claude desktop app's Code tab is the easiest start, with no terminal. If you do use the terminal, this page walks through opening PowerShell on Windows or Terminal on a Mac, the install command, and how to open your project folder.

Do I run the install command every time?

No. You install once. After that, open the desktop app, or open a new terminal, go to your project folder and type claude.

Do I need to know how to code?

No, not to start. You describe the job in plain English. You do need to read what it plans to do, approve permission prompts, and check the result before you share or publish it.

Can it break my project?

It can change and delete files you allow it to touch. Work in a copy, ask it to plan first, and use rewind or git so you can go back.

What do I do when I don't understand an error?

Paste the message back and ask it to explain in plain words, what it was trying to do, and the safest next step. Don't approve a fix you can't follow.

Related: Claude Code explained: what it is and how to set it up · Control Claude Code from your phone or the web · AI agents, explained for normal people · Get more out of your Claude plan for less · Build a mini-app with AI, no coding · Claude plans and prices

Sources (15)

  1. Claude Code docs: Get started with the desktop app
  2. Claude Code docs: Terminal guide for new users
  3. Claude Code docs: Choose a permission mode
  4. Claude Code docs: Security
  5. Claude Code docs: Checkpointing
  6. Claude Code docs: Advanced setup
  7. claude.com/pricing
  8. Claude Code docs: Authentication
  9. Claude: Download
  10. Claude Help Center: Install Claude Desktop
  11. Claude Code docs: Quickstart
  12. Claude Code docs: How Claude remembers your project
  13. Claude Help Center: Models, usage and limits in Claude Code
  14. Claude Code docs: Projects
  15. Claude Code docs: Remote Control

Last checked Nov 14, 2026.

Prices change often. Get the changes by email.

New models, price changes and the best deal of the week. One email on Fridays. Free.

Sign up for weekly AI alerts. We'll send the next issue when it goes out. Sign-up opens soon. Until then the button opens your email app with a pre-filled request to [email protected], and we'll add you when the next issue goes out.

AI moves fast. We'll keep you up to speed.

New models, price changes and the best deal of the week. One email on Fridays. Free.

Sign up for weekly AI alerts. We'll send the next issue when it goes out. Sign-up opens soon. Until then the button opens your email app with a pre-filled request to [email protected], and we'll add you when the next issue goes out.