Claude Code for Beginners: 10 Steps to Get Started (2026)

by Jasmine, a dev who uses Claude Code every day.

I wrote this for the exact person I was a little over a year ago: someone who felt a jolt of fear opening a terminal, had no idea what to type, and worried that every AI run would torch a pile of token money. Good news - Claude Code is far easier to start with than you think. If you want to understand the fundamentals first, skim what Claude Code is (an A-Z explanation); if you just want "follow along and it works," jump straight into the 10 steps below.

Before you start: what you need

A few things below are all it takes. Do not worry if you are shaky with the command line - every step spells out exactly what to type.

Prep checklist (5 minutes):

• A computer running Windows, macOS, or Linux - any of them works.
• Node.js (the LTS build) - needed for the npm install method. Check it with node -v; if that errors out, download and install Node.js LTS first.
• A Claude account: either the Pro plan at $20/month (the easiest and cheapest option for beginners) or an API key billed per token.
• A project folder to experiment in - a completely empty folder is fine.
• About 30 minutes and a "let's just try it" mindset.

You do not need to be good at the terminal. You only need two commands: cd to move into a folder and claude to open it. Everything else you say in plain English.

Step 1 - Install Claude Code

The most common way is to install via npm. Open a terminal (on Windows, use PowerShell or Terminal) and type:

npm install -g @anthropic-ai/claude-code

This installs Claude Code globally, so you can call claude from any folder (per the official docs). Once it finishes, verify it with:

claude --version

OS-specific notes:

• Windows: if typing claude gives you a "command not recognized" error, close the terminal and reopen it to reload your PATH. Running PowerShell is your safest bet.
• macOS/Linux: if you hit a permissions error on the global install, avoid sudo - instead, reconfigure your npm global directory, or use a Node version manager.

I deliberately kept every edge case out of this section so the article stays readable. If you get stuck, see the detailed Claude Code installation guide for each platform and how to fix the errors people run into most.

Step 2 - Sign in and authenticate

The first time you open it, Claude Code asks you to authenticate. There are two routes:

My advice for beginners: pick the Pro plan at $20/month. You do not have to track every token, you will not "accidentally" spend a lot while you are still learning, and the fixed price is easy to reason about. Move to an API key when the work gets bigger. Just type claude, choose your sign-in method, and follow the on-screen prompts - it takes a few dozen seconds.

Step 3 - Open your first project and run /init

Move into your project folder and start Claude:

cd path/to/your-project
claude

When the chat interface appears in the terminal, type:

/init

The /init command tells Claude to scan your project and create a CLAUDE.md file with a high-level overview (folder structure, tech used, conventions). This file helps Claude "understand the context" in every later session - you will fine-tune it in Step 8.

Step 4 - Understand the basic loop

Before you give any instructions, grasping this loop means you understand 80% of how Claude Code works. It repeats like this:

1. You give a prompt - say in plain English what you want.
2. Claude reads the repo - opens the relevant files to understand the context.
3. Claude proposes changes - shows you which files it will edit or create.
4. You approve (permission) - Claude asks before writing a file or running a command; you accept or decline.
5. Claude edits files and runs tests - then reports the result.
6. Repeat until it matches what you wanted.

About permissions: by default Claude always asks first before any action with impact - this is the "safety brake" for beginners. Read carefully, then approve. Do not rush to turn on auto-accept-everything before you are comfortable (it is one of the 5 common mistakes at the end of this article). Keep the habit: read, understand, then approve.

Here is what puts a lot of beginners at ease: Claude does not quietly delete files or run dangerous commands behind your back. Every time it wants to write a file, run a test, or call a terminal command, you see exactly what it is about to do and you can decline. In other words, you are always the one driving - Claude only proposes and then executes when allowed. Once you understand this loop, every remaining step is just a variation on it.

Step 5 - Write your first prompt the right way

The quality of the result depends directly on how you phrase your instructions. The golden rule for beginners: small, specific, one thing at a time. Do not ask Claude to "build the whole website" in a single sentence - break it down.

A good prompt always has context (which file, which line), a clear goal, and a scoped boundary. The more specific you are, the less Claude guesses and the less you have to redo.

Step 6 - Use Plan Mode before you build

For anything even slightly complex, do not let Claude dive straight into code. Ask it to plan first. My mantra: one minute reading the plan saves ten minutes building in the wrong direction - and it saves tokens too, because you are not redoing the work.

Step 7 - Your first real project (with actual output)

Now for the fun part: building something you can actually see. I picked an extremely light mini-project so anyone can follow along - create a static HTML landing page. You can apply the same process to "fixing a small bug" if you already have a project.

How to do it: open a terminal in an empty folder, type claude, then give a specific prompt (remember the rule from Step 5):

Create an index.html file: a static landing page for a coffee shop called "Sunny".
Include: a header with the shop name, one slogan, a "Menu" section listing 3 items with prices,
and a footer with the address. Use clean inline CSS with basic responsiveness.
No JavaScript needed yet.

Claude will propose creating index.html, ask permission to write the file, and you approve. Once the file is created, open it in a browser (double-click index.html or drag it into a browser tab) to see the real rendered result.

Honesty check: the first pass usually nails the skeleton but may not match your taste - pale colors, say, or spacing that is not quite right. Do not get discouraged: just keep talking, "Give the Menu section a cream background, round the corners of each item, and add more space between them." This is exactly the loop from Step 4. What Claude is really good at is scaffolding fast and adjusting to short requests; where you need to stay active is spelling out your aesthetic and checking the result in the browser. After a few rounds, you have a finished page without hand-writing a single line of HTML.

Step 8 - Create a CLAUDE.md for your project

CLAUDE.md is a text file at the root of your project that records what you want Claude to always remember: what this project is, which tech it uses, how the code is styled, and the test/run commands. Every new session, Claude reads this file first, so you do not have to repeat yourself from scratch.

Beginners should have a CLAUDE.md early because it drastically cuts down on Claude "losing context" and guessing wrong. A minimal CLAUDE.md can look like this:

# Project: Sunny coffee shop landing page

## Tech
- Static HTML/CSS, no framework, no build step.

## Conventions
- Keep CSS inline in index.html for simplicity.
- Write display content in English.

## Common tasks
- Adjust layout, colors, menu content.

You already have a draft from /init back in Step 3 - now you just open it and edit it to match reality. To write a CLAUDE.md properly (structure, tips, examples for larger projects), see the guide to writing a solid CLAUDE.md for your project.

Step 9 - Learn the essential shortcuts and slash commands

You do not need to memorize them all. Knowing about 6 core commands is enough for daily use:

Small tip: use /clear whenever you start a task unrelated to the previous one - it is the most effective token-saving habit for beginners. For a fuller table, see the complete Claude Code shortcuts and commands cheat sheet. Once you are comfortable, you can also add prebuilt skills and slash commands so you stop retyping the same repetitive workflows - which is exactly what Step 10 is about.

Step 10 - Speed up with prebuilt skills and subagents

After a few projects run smoothly, you will notice you repeat a lot: the same kind of code-review prompt, the same backend-scaffolding process, the same way of writing tests. This is where skills and subagents shine - they are "prepackaged workflows" for Claude to follow, instead of you describing them again every time.

Beginners are absolutely not required to build everything themselves. There are kits that bundle a whole range of skills and agents for both coding and marketing - for example, the AgentKit kit for Claude Code (detailed review). For beginners, I treat this as an upgrade once you are comfortable, not something you need on day one: get the 9 steps above solid first. When you find yourself losing time retyping the same workflow, give AgentKit a try (20% off via link) to see how much effort a prebuilt skills/subagents system can save. Do not rush to buy before you actually need it - that is the honest rule.

5 common beginner mistakes (and how to avoid them)

These are the mistakes I see most in beginners. Avoiding them alone makes you noticeably faster:

• 1. Assigning a task that is too big. "Build the whole app for me" makes Claude guess and wander. How to avoid: break it down, one clear thing per prompt (see Step 5).
• 2. Skipping Plan Mode. Diving straight into code easily heads the wrong way and forces a redo, wasting both time and tokens. How to avoid: ask for a plan first on anything touching more than one file (Step 6).
• 3. Auto-accepting everything without reading. Turning on auto-accept and letting Claude edit at random. How to avoid: keep the read-understand-then-approve habit for permissions, at least until you are truly comfortable.
• 4. Not using /clear. Cramming every task into one session dilutes the context, weakens the answers, and raises token use. How to avoid: run /clear whenever you switch tasks.
• 5. No CLAUDE.md. Without this file, Claude has to re-guess your project every session. How to avoid: create a CLAUDE.md early and update it over time (Step 8).

Frequently asked questions (FAQ)

Is Claude Code free?

There is a limited trial tier, but for steady use you need a Claude plan (Pro at $20/month is the popular pick for individuals) or pay-per-token via an API key. For beginners, the Pro plan is the easiest to keep costs predictable because the price is fixed.

Do I need to know how to code first?

Not required to get started - you give instructions in plain English and Claude writes the code. But knowing the basics (reading code, understanding what your project contains) helps you write better prompts and check the results. The more you learn, the more effectively you use Claude Code.

Does it run on Windows?

Yes. Claude Code runs on Windows, macOS, and Linux. On Windows, use PowerShell or Terminal; if typing claude is not recognized, close and reopen the terminal to reload your PATH.

How much do tokens/money cost?

On the Pro plan at $20/month the cost is fixed, so you do not have to worry about individual tokens while learning. On an API key you pay by token volume (for example, Sonnet 5 is roughly $2 in and $10 out per 1 million tokens). Using /clear when you switch tasks and turning on Plan Mode noticeably reduce token use.

How is it different from ChatGPT or Claude chat?

Regular chat only answers in text, and you have to copy code back and forth yourself. Claude Code runs right in your terminal, reads your project directly, and edits files and runs commands (after asking permission). In short: chat is an assistant that answers, Claude Code is a teammate that operates on your actual codebase.

How long does it take to learn Claude Code?

Getting your first project running takes only about 30 minutes, as in the 10 steps above. Getting proficient (good prompts, context management, using skills/subagents) usually takes a few days of practice. Just build real projects and you get used to it fast.

Conclusion and next steps

With these 10 steps done, you have gone from zero to a first working project - install, authenticate, understand the loop, write good prompts, and avoid the common mistakes. That is a solid foundation to go further. My suggested next steps: bookmark the complete commands and shortcuts cheat sheet for quick reference, read up on how to write a solid CLAUDE.md, and if you want to understand the fundamentals deeply, circle back to what Claude Code is (an A-Z explanation).