Claude Code Tutorial for Beginners: Your First Session

Claude Code Tutorial for Beginners

Claude Code is a coding assistant that lives in your computer’s terminal (or in a desktop app) and works directly inside your project folder. Unlike a chat window, it can read your files, change them, and run commands, and it asks your permission before doing the risky parts. You can start with no coding background, as long as you begin with a small project and read what it changes. The first step is simple: make or pick a small folder, open it, and ask Claude Code to explain what’s inside without changing anything.

That one habit, looking before touching, is what separates a good first session from a messy one. This guide walks through install, sign-in, and a first session built around five prompts, then covers the controls that keep you safe.

If you’ve never described a program in plain language before, the basics in this beginner guide to vibe coding make the rest of this much easier to follow.

I went through Anthropic’s setup documentation, its permission-mode guidance, and its official cheatsheet in October 2026 for the details below. Tools like this change fast, so the examples are marked as illustrations, and anything that depends on a plan or price is something to confirm on the official page the day you sign up.

What Claude Code Actually Is

A chatbot answers you, and you copy the result into your files yourself. Claude Code works the other way around. You describe a goal, and it explores your project, decides which files matter, makes changes, and can run commands to check its own work. It’s an agent: a tool that picks its own steps instead of waiting for each one.

That makes it powerful and also worth handling carefully. It acts on your real files, so the habits that matter are the same ones a careful person would have with a new assistant: start small, check the work, and keep a way to undo.

Two terms before we go on. The terminal is the text window where you type commands instead of clicking. A project folder is just the folder that holds your code. Claude Code works inside one project folder at a time.

What You Need Before You Start

You need three things: a paid way in, a computer that can run it, and a small project to practise on.

A paid way in

The free Claude plan doesn’t include Claude Code. You need a paid Claude plan (Pro, Max, Team, or Enterprise) or a Claude Console account, which bills by usage. Paying usually means an international card, which is a hurdle many people already know from other subscriptions, so check what the billing page accepts before you plan around it. Plans and prices change, so confirm the current ones on Anthropic’s pricing page.

A computer that can run it

macOS and Linux work, and so does Windows 10 (version 1809 or later) or newer, either natively or through WSL, which is a way to run Linux inside Windows. It needs an internet connection, since the thinking happens on Anthropic’s servers. That also means the project text you work on is sent there, which matters if you’re handling someone else’s confidential code.

A small practice project

One folder with a few files is plenty. A simple web page with an HTML file, a CSS file, and a JavaScript file is ideal.

There’s one billing trap worth knowing before you install anything. If an API key is stored on your computer, Claude Code can use it instead of your subscription. Anthropic’s documentation lists a stale key overriding a subscription login as a known cause of authentication problems, and users have reported ending up billed per use without realising. If you’ve ever worked with an API key, check your environment and your project’s .env file before your first session.

Pick Your Surface: Terminal, Desktop App, or Editor

Claude Code runs in the terminal, in editors like VS Code, in a desktop app, and on the web. For a beginner, three choices matter.

If typing commands makes you nervous, use the desktop app. It lets you use Claude Code without the terminal, and it’s available for macOS and Windows.

If you want to learn how developers actually work, use the terminal. It’s the original and most complete way to run it, and the commands you’ll learn carry over everywhere.

If you already live in VS Code, use the editor extension, so Claude Code sits next to your files.

My suggestion: if you’re unsure, try the terminal for your first project. It looks scarier than it is, and the steps below are short. The official setup docs list every install option and the system requirements if you want to check anything against the source.

Install and Sign In

The recommended method is the native installer. It needs no extra software, and it updates itself in the background.

On macOS, Linux, or WSL, open a terminal and run:

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

On Windows, open PowerShell and run:

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

On native Windows, Git for Windows is recommended so Claude Code can use its Bash tool. If it isn’t installed, Claude Code falls back to PowerShell.

When it finishes, close and reopen the terminal, then check that it worked:

claude --version

If something looks wrong, claude doctor runs a health check and points to the problem.

Now go to your practice folder. On Windows 11 you can right-click inside the folder and choose “Open in Terminal.” Otherwise, type cd followed by the folder path. Then start Claude Code:

claude

The first time, it asks you to sign in. A browser window opens, you log in to your Claude account, and you come back to the terminal. After signing in, type /status and look at the login method. It should show your subscription account. If it shows an API key instead, stop and fix that before you use it for real work, because that’s the billing trap from earlier.

Permission Modes: How Much Freedom to Give It

Before the first prompt, learn the one control that matters most. Claude Code has permission modes, which decide how often it stops to ask before acting. You switch between them with Shift+Tab.

In default mode, it asks before editing files or running commands. This is where you should start.

In accept edit mode, it stops asking about file changes for the rest of the session, but it still asks before running commands. It’s useful once you trust the direction of the work.

In plan mode, it’s read-only. Claude explores, explains, and proposes a plan, but it won’t edit files or run commands. Press Shift+Tab to cycle, or start a single prompt with /plan.

This is the same “review before it acts” idea that runs through the guide on AI agents, and plan mode is the cheapest safety tool you’ll get. There are riskier modes that skip prompts entirely. As a beginner, leave them alone.

Your First Session in Five Prompts

Do these in order, in your small practice folder. Each one builds on the last.

Prompt 1: Look, don’t touch. Start in plan mode or just say it outright:

Look through this folder and explain what this project does and how the files fit together. Don’t change anything.

Read the answer. If it doesn’t match what you know about the project, you’ve learned something important before any damage was possible.

Prompt 2: Plan before you change. Pick one small improvement, like a dark-mode button, and ask for a plan:

I want to add a dark-mode toggle to the page. Before changing anything, tell me which files you’d edit and what you’d do in each.

Check the plan. If it wants to touch ten files for a button, say so and ask for a smaller approach.

Prompt 3: One change only. Switch back to default mode and say:

Go ahead and make that change, but only that change. Don’t tidy or restructure anything else.

It will show each edit and ask to approve it. Read each one. This is where you learn the most.

Prompt 4: Check it. Ask it to prove the change works:

Run the project’s tests if there are any. If not, tell me how to check this by hand, and show me what you changed.

Then open the page yourself and click the button. Don’t accept “it works” without looking.

Prompt 5: Save and summarise. Ask for a plain-language summary of what changed. If the folder uses Git, a version-control tool that saves snapshots of your project, ask it to commit the change with a clear message. If you’ve never used Git, ask Claude Code to explain it and set it up. Having a snapshot means you can always go back.

A few commands help along the way. /init scans the project and writes a starter CLAUDE.md file (more on that next). /clear starts a fresh conversation, /compact shrinks a long one, and /rewind goes back to an earlier checkpoint. Pressing Esc twice opens the same rewind menu. Anthropic’s official cheatsheet lists the rest, and it’s worth keeping open for your first week.

Checkpoints undo what Claude changed in the current session, but they aren’t a replacement for Git. Think of rewind as session-level undo and Git as your real save file.

CLAUDE.md: Teach It Your Project Once

CLAUDE.md is a plain text file in your project folder that Claude Code reads at the start of every session. It’s where you write the things you’d otherwise repeat: how to run the project, what style you prefer, and what it should never touch.

Run /init and Claude Code writes a first version for you. Then edit it. Keep it short and specific. For example:

This is a small website. Open index.html in a browser to test it. Never edit the files in the /backups folder. Ask before installing anything new.

If a rule matters, put it here. And keep secrets out of the project folder, since Claude Code can read what’s in it.

Three Beginners, Three Projects

These are illustrative scenarios based on how these tools usually behave. They are not reports about real people.

The student with a portfolio page

He has a small site made of three files. He starts with prompt 1 to understand a template he downloaded, then asks for a contact form layout. Explaining the files went smoothly, and so did small edits. It stalled when he typed “make it look modern.” That one request touched most of his files. Now he uses plan mode first, and asks for one section at a time.

The freelancer fixing a client’s page

The mobile menu on a client’s landing page won’t open. She asks Claude Code to find the cause, and it does quickly. It stalled when the fix came bundled with tidy-ups to unrelated code. She now names the files it may touch, reads every edit, and commits the fix on its own so the client can see exactly what changed.

The beginner with a downloaded project

She has a folder of code she doesn’t understand and uses Claude Code purely as a guide, asking questions in plan mode. It went smoothly for explanations. Her problem was usage: very long sessions used up her plan allowance. She learned to start a fresh conversation for each topic and to point at specific files instead of the whole folder. She also asks what a script does before running anything she didn’t write.

Keeping Usage Under Control

Claude Code keeps your conversation in a working memory called the context window. As it fills, older content gets summarised or dropped, answers can get vaguer, and you use more of your plan. Plans have a rolling allowance, so long, messy sessions burn through it faster.

Four habits help. Start Claude Code from your project folder, not your whole drive. Use /clear between unrelated tasks. Use /compact when you’re still on the same task but the chat has grown long. And point to specific files instead of asking it to read everything.

Mistakes That Cost Beginners Time

Starting in the wrong folder. If you launch it in your home folder, it can see far more than it should. Always start inside one project.

Approving commands without reading them. The prompt shows exactly what it wants to run. Read it, especially anything that deletes, installs, or downloads.

Switching to accept-edits too early. Get a feel for the edits first. Then give it more freedom.

Skipping Git. Rewind helps, but a real snapshot is safer.

Asking for a whole app in one prompt. Big requests produce big, hard-to-check changes. Ask for one piece at a time.

Trusting “it works.” Run it yourself. Open the page, click the button.

Ignoring the login check. An API key quietly overriding your subscription is an expensive way to learn.

Three Quick Self-Checks

The install check. Run claude –version, and if anything seems off, claude doctor. If the command isn’t found, close and reopen the terminal.

The read-only check. Start in plan mode, ask a question about your project, then check that no files have changed. This proves plan mode does what it says.

The undo check. Make a tiny change, such as a comment in a file, then press Esc twice and rewind. Seeing undo work once makes you braver, in a good way.

Decision Checklist: Is Claude Code the Right Tool for You?

Answer yes or no to each one.

  • Can you pay for a plan or Console credits with a method your bank allows?
  • Are you comfortable typing a few commands, or willing to use the desktop app instead?
  • Do you have a small project in a folder, ideally with a backup or Git?
  • Are you willing to read each change before approving it?
  • Is your goal to learn and build small pieces, not to get a finished app with no effort?
  • Do you have a stable internet connection?
  • Is it acceptable for your project’s text to be processed on Anthropic’s servers?

If you answered yes to most, start with the five prompts. If the first question is a no, there are other ways in, but not a free one inside Claude Code itself. If the last one is a no, don’t use it on that project.

When Claude Code Is Not the Right Choice

It’s the wrong tool if you can’t pay for any plan, if a client’s contract forbids sending their code to outside services, or if you need to work fully offline. In those cases, a smaller local model or a plain chat window with a reviewed copy-and-paste is wiser.

It’s also not ideal for a quick, one-off question. A normal chat is faster and cheaper than opening a project session. And it’s not a substitute for learning the basics. If you can’t read a single line of the code it writes, use it as a teacher first, asking it to explain everything, until you can follow along.

Features and plans change often, so before you commit, skim Anthropic’s official quickstart to see what’s current.

Quick Problem Diagnosis

What you seeLikely causeFirst thing to check
“Not logged in”Sign-inType /login and sign in again.
“Credit balance too low” but you have a planWrong login, usually an API key overriding itCheck /status and any API key in your environment or .env file.
It edits files you didn’t expectThe prompt or scopeName the files it may touch, use plan mode, and read the changes.
Answers get vague late in a sessionA full context windowRun /compact, or /clear and start fresh.
“claude: command not found”Install or PATHReopen the terminal, then run the installer again or claude doctor.
It keeps asking for permissionThe permission modeCheck which mode you’re in with Shift+Tab.
“Usage limit reached”Plan allowanceWait for the reset, or shorten sessions and use /clear between tasks.

Questions Beginners Ask

Is Claude Code free? 

Not on the free Claude plan. You need a paid plan or a Console account billed by usage. If you can’t pay, a local model is a free but slower and weaker alternative, and this guide to local DeepSeek shows what’s realistic on an ordinary PC.

Do I need to know how to code? 

No, but you do need to read what it changes. Start with small projects and ask it to explain every edit in plain words.

Is it safe to let it run commands? 

By default it asks first, and you can read each command before approving it. Stay in default or plan mode while you learn, and work in one project folder.

Should I use the terminal or the desktop app? 

Use the desktop app if the terminal feels like too much. Use the terminal if you want to learn the tool that developers use every day.

How is it different from chatting with Claude in a browser? 

A chat window gives you text to copy. Claude Code works inside your folder, reading and editing real files and running commands, with your permission.

Will it work on a slow connection? 

It needs a connection, but it mostly sends and receives text, so it’s lighter than video or image tools. A very unstable connection will cause interruptions.

What if international card payments are hard where I live? 

Check which payment methods the billing page accepts before you plan around it. If none work for you, try a local model or a plain chat assistant in the meantime.

Your First Hour

Make a small practice folder, install Claude Code, sign in, and check the login method. Run the five prompts in order, saving a snapshot after prompt 5. Then try the undo check once.

After that, build something slightly bigger, one piece at a time. The people who get the most from this tool aren’t the ones who ask for the most. They’re the ones who check every change, so the tool can keep earning more trust, one small task at a time.

Spread the love

Similar Posts

Leave a Reply

Your email address will not be published. Required fields are marked *