Claude Code for Beginners: Install, Costs and First Steps

Yes, a non-developer can learn Claude Code in an afternoon, and the fastest route is the desktop app with a paid Claude plan, not the terminal. This guide to Claude Code for beginners is written for founders and operations leads who want an AI agent that reads files, writes scripts and fixes small things on their own machine, without first becoming a programmer. You will get the exact install commands, the first ten minutes of a session, what it costs, which settings keep you safe, and the habits that separate useful sessions from expensive ones.

Everything below comes from Anthropic’s own documentation and pricing page, checked on November 4, 2026. Claude Code changes quickly, so each section links to the source page you should re-check before you rely on a detail.

What Claude Code is, in plain terms

Claude Code is an AI agent that works inside a folder on your computer. You describe a task in plain English. It reads the files in that folder, proposes changes, runs commands, and shows you what it did. A chat assistant gives you text to copy and paste; Claude Code does the work in place and asks your permission along the way. Anthropic’s free Claude Code 101 course frames the difference the same way: a coding agent runs a loop of reading, acting and checking, while a chat tool only answers.

That loop is why it is useful for people who are not engineers. An ops lead can point it at a folder of CSV exports and ask for a cleaned, merged file. A founder can ask it to build a simple internal web page, rename a folder of screenshots by their content, or explain what a contractor’s script does before running it. The official terminal guide for new users lists exactly these kinds of first tasks, including renaming the screenshots on your Desktop based on what each image shows.

It is not magic, and it is not free. Claude Code needs a paid Claude subscription or API credits, it can make mistakes you will have to catch, and it does its best work when you give it specific instructions. The rest of this guide is about getting those three things right from day one.

Where you can run it

Claude Code is one product with several front doors. The quickstart lists the terminal, the web at claude.ai/code, a desktop app, VS Code and JetBrains extensions, Slack, and CI pipelines such as GitHub Actions. All of them run the same agent. For a beginner the choice is mainly about comfort: do you want a window with buttons, or a text prompt?

Who should start with Claude Code, and who should not

Claude Code is a good fit if you already have files or a small project you want to change, and you are willing to read what the agent proposes before approving it. It suits operations people who live in spreadsheets and exports, founders who maintain a simple website or internal tool, and agency buyers who want to understand what their developers are shipping. It also suits anyone who wants to automate a repetitive computer task without paying for a custom build first.

It is a weaker fit in three cases. First, if you want a finished app with hosting, logins and a database handled for you, an app builder is usually faster, because Claude Code edits files on your machine and leaves deployment to you. Second, if your work is mostly writing, research and documents rather than files and scripts, Anthropic’s Cowork mode is closer to what you need; our comparison of Claude Code vs Cowork walks through that split. Third, if you cannot review changes at all, you will not know when the agent is wrong, and that is where the real risk sits.

A useful test: if you can describe the result you want in two or three sentences and you could check it with your own eyes (a file opens, a page loads, a total matches), you are ready. If you cannot describe the result or check it, start with a smaller task until you can.

Pick your starting point: desktop app, terminal, VS Code or web

The single most useful decision for a beginner is which interface to use first. All four options below need the same paid plan, and you can switch later without losing anything, because your project files stay where they are.

Sources for the table: the desktop quickstart, the system requirements page, and the VS Code extension docs. The desktop docs say plainly that the app “includes Claude Code, so you don’t need to install Node.js or the CLI to use the Code tab.” That is the main reason we recommend it for week one: there is nothing to configure, and every change shows up as a visual diff.

Move to the terminal once you have a few successful sessions behind you. The terminal gives you scripting, one-off commands such as claude -p "explain this function", and the full set of options. If you are comparing Claude Code against editor-first tools, our Antigravity vs Cursor vs Claude Code breakdown covers where each one is stronger.

How to install Claude Code, step by step

Before anything else, buy a plan. The setup page is explicit that “the free claude.ai plan does not include Claude Code access,” so a Pro, Max, Ultra, Team, Enterprise or Console account is required. Then follow the path for your interface.

Path A: the desktop app (no terminal)

1. Download the Claude app for macOS or Windows from the links on the desktop quickstart, or install it with apt on Ubuntu and Debian.
2. Open it and sign in with your Anthropic account.
3. Click the Code tab at the top of the sidebar. If it asks you to upgrade, your account is still on the free plan.
4. Under the prompt box, choose the Local environment, then Open folder… and pick a small project you know well.
5. Choose a model from the picker and type your first request.

Path B: the terminal

Open a terminal. On a Mac, press Cmd + Space, type Terminal and press Enter. On Windows, press Win + X and choose Windows PowerShell or Terminal. Then paste the command for your system, taken from the official quickstart:

# macOS, Linux, WSL
curl -fsSL https://claude.ai/install.sh | bash

# Windows PowerShell (prompt starts with PS C:\)
irm https://claude.ai/install.ps1 | iex

# Windows CMD (prompt starts with C:\ and no PS)
curl -fsSL https://claude.ai/install.cmd -o install.cmd && install.cmd && del install.cmd

The installer shows no progress bar, so wait until it finishes. Then open a new terminal window and check the install:

claude --version
claude doctor

The first command prints a version number followed by “(Claude Code)”. The second prints read-only diagnostics about the install and your settings without starting a session. If your shell says claude is not found, the install folder is not on your PATH yet. On a Mac with the default Zsh shell, the terminal guide gives this fix:

echo 'export PATH="$HOME/.local/bin:$PATH"' >> ~/.zshrc
source ~/.zshrc

On Windows, installing Git for Windows is optional but recommended: it gives Claude Code a Bash shell, and without it Claude Code falls back to PowerShell. You can accept every default screen in that installer.

Alternative installers

Homebrew (brew install --cask claude-code) and WinGet (winget install Anthropic.ClaudeCode) also work, but per the setup docs they do not auto-update, while the native installer updates itself in the background. Beginners should use the native installer for that reason alone. If you see older guides recommending npm install -g, note that the npm package now needs Node.js 22 or later and should never be run with sudo.

Your first session in ten minutes

Start in a folder you can afford to break. Copy a real project into a new folder, or ask Claude Code to create a tiny one. In the terminal, move into the folder and start a session:

cd ~/Documents/claude-practice
claude

The first time, a browser window opens so you can log in. After that, the quickstart suggests starting with questions rather than changes, because questions cost little and show you how the agent thinks:

what does this project do?
explain the folder structure
where is the main entry point?

Then try one small change, for example make me a simple webpage that says hello world. When Claude asks before creating or changing files, read the proposal and choose Yes. Open the result yourself (double-click the HTML file) and confirm it does what you asked. That last step is the habit that matters most.

The handful of commands you need

You can ignore most commands at first. These are the ones beginners actually use, from the quickstart and checkpointing docs:

• /help lists available commands. Typing / on its own shows commands and skills.
• /clear wipes the conversation so the next task starts fresh.
• /rewind, or pressing Esc twice on an empty prompt, opens a menu to restore code, conversation or both to an earlier point.
• /usage shows how much of your plan you have used.
• claude -c continues the most recent conversation in the current folder.
• Esc stops Claude mid-task. /exit or Ctrl + D twice leaves the session.

Give Claude Code a memory with CLAUDE.md

Every session starts with a blank slate. The fix is a plain text file called CLAUDE.md in your project folder. Claude Code reads it at the start of every session, so anything you write there (what the project is, how to run it, what never to touch) does not have to be repeated. The memory documentation recommends running /init to generate a starting file: Claude scans the folder and writes down the commands and conventions it finds. If a CLAUDE.md already exists, /init suggests improvements rather than overwriting it.

Keep it short. Anthropic’s guidance is to “target under 200 lines per CLAUDE.md file,” because longer files use more context and are followed less reliably (source). A beginner’s CLAUDE.md for a folder of business data might be ten lines:

# Sales exports project
- Raw exports live in /raw. Never edit or delete files in /raw.
- Write all outputs to /clean as CSV with a header row.
- Dates are DD.MM.YYYY in the raw files; convert to YYYY-MM-DD.
- Currency is EUR unless the column name says otherwise.
- Before finishing, print row counts for input and output.

There are three scopes worth knowing: ./CLAUDE.md for the project (shared with anyone who uses the folder), ~/.claude/CLAUDE.md for your personal preferences across all projects, and CLAUDE.local.md for private project notes you do not want to share. Claude also keeps its own “auto memory” notes from your corrections, which you can toggle in /memory. This repository, the one that publishes this blog, runs on the same idea; the guide to building an AI agent with Claude Code shows how far a single CLAUDE.md plus a few skills can go.

Permission modes: how much the agent can do without asking

The setting that most affects safety is the permission mode. It decides which actions Claude takes on its own and which ones wait for your approval. Press Shift + Tab to cycle modes during a session, or start in a specific mode with a flag. The permission modes reference lists six:

One detail catches beginners out. The same reference notes that from version 2.1.283, terminal and VS Code sessions start in Auto by default. Auto blocks obviously dangerous actions, but the docs are clear that it “reduces prompts but doesn’t guarantee safety.” For your first sessions, start explicitly in a stricter mode so you see every step:

claude --permission-mode plan

Plan mode is the best teacher. Claude reads the folder, writes out what it intends to do, and waits. You learn how it approaches a problem, and you can correct the plan before a single file changes. That is also why the cost guide recommends plan mode for complex tasks: it prevents paying for work in the wrong direction.

What Claude Code costs a beginner

There is no separate Claude Code price. You pay for a Claude plan, and Claude Code draws on that plan’s usage. Prices from claude.com/pricing, checked November 4, 2026:

The alternative is pay-per-token billing through a Claude Console account. Anthropic’s cost documentation reports that across enterprise deployments the average is “around $13 per developer per active day and $150-250 per developer per month,” with 90% of users under $30 per active day. Those are full-time engineers, so treat them as a ceiling, not a beginner’s bill.

A worked example for a small ops team

Say three people in operations want to try Claude Code for cleaning exports and building small internal tools. Using the published plan prices, three Pro subscriptions billed annually cost 3 x $17 = $51 per month. Three Team standard seats billed annually cost 3 x $20 = $60 per month and add central billing and admin controls. Now compare API billing: if each person had 10 active days a month at the enterprise average of $13 per active day from the cost guide (our assumption for light use), that is 3 x 10 x $13 = $390 per month. For interactive, part-time use, a subscription is the cheaper and more predictable choice. API billing makes sense later, for scheduled jobs that run without a person at the keyboard.

The limit you will feel on Pro is the five-hour usage window, not money. If you hit it often, move one heavy user to Max rather than upgrading everyone. Before that, check /usage: on paid plans it attributes usage to skills, plugins and MCP servers and recommends specific changes, such as trimming an oversized CLAUDE.md.

Beginner mistakes that waste money or break things

Most bad first experiences come from a short list of habits. Anthropic’s best practices page and the cost guide name most of them directly.

• Vague requests. “Improve this project” makes Claude scan everything and guess. Name the file, describe the symptom, and say what “done” looks like. The best practices page notes that a specific prompt “usually save[s] yourself a round of corrections later.”
• One endless conversation. Claude re-sends the whole conversation with every request, so a session open all day keeps paying for its history. Run /clear between unrelated tasks; Anthropic notes it “costs nothing.”
• Correcting the same thing a third time. The guidance is blunt: after two failed corrections, run /clear and write a better prompt that includes what you learned, instead of piling on.
• Trusting rewind as a backup. /rewind restores files Claude edited with its own tools. Per the checkpointing docs, a file changed only by a shell command is not tracked, and checkpoints are “not a replacement for version control.” Keep a copy of anything important, or ask Claude to set up Git for the folder.
• Accepting without checking. Generated code can look right and still be wrong. Open the output, check a total, load the page. If there is a test or a known-good number, give it to Claude up front so it can verify its own work.
• Running in the wrong folder. Claude can read and edit what is inside the folder you start it in. Starting it in your home folder gives it far more reach than a beginner needs.

A real first project: cleaning a monthly sales export

Here is a typical first-week task for an operations team, written as the prompts you would actually type. The folder has a raw subfolder with twelve monthly CSV exports from a store platform, each with slightly different column names.

Step 1, explore in plan mode. Start with claude --permission-mode plan and ask: look at the files in raw/ and tell me how their columns differ. Don't change anything. Claude lists the variations, for example one month that calls the field “Order Total” and another that calls it “total_eur.”

Step 2, agree the plan. Ask: propose a plan to merge all twelve files into clean/sales-2026.csv with columns date, order_id, customer_email, total_eur. Convert dates to YYYY-MM-DD. Keep raw/ untouched. Read the plan. If it wants to delete duplicates, decide whether that is right for your data and say so.

Step 3, approve and verify. Approve the plan, switch to Manual or Accept edits with Shift + Tab, and let it write a small script and run it. Then ask: print the row count per input file and the total row count in the output, and list any rows you dropped and why. If the counts do not add up, you have caught the problem before anyone used the file.

Step 4, save the knowledge. Run /init or ask Claude to add the column rules to CLAUDE.md, so next month’s run takes one sentence. Once a task like this works reliably, it becomes a candidate for a scheduled job, which is where Claude Code starts to look like the broader Claude-based automation businesses run in production.

When another tool fits better than Claude Code

Claude Code is the right first agent for file and script work on your own machine. It is not the only option, and a beginner should know when to pick something else.

• Cowork, in the same Claude app, suits document, research and desktop tasks that do not involve a code project. The plan you pay for covers both.
• Editor-first tools such as Cursor suit people who want to see and type code themselves with AI assistance alongside, rather than delegating whole tasks.
• App builders suit founders who want a hosted app with logins and a database and do not want to manage files or deployment at all.
• Workflow tools such as n8n, Make or Zapier suit recurring business processes that connect SaaS apps on a schedule. Claude Code can help you build or debug those workflows, but the workflow tool is what should run them every day.

The rule of thumb: if the work lives in files on your computer and you want an agent to do it, start with Claude Code. If the work lives in other people’s SaaS tools and repeats on a schedule, use a workflow tool and bring Claude Code in to build it.

Claude Code beginner checklist

Run through this list before and during your first week. Each item maps to a section above.

1. Paid plan active (Pro is enough to start); the free plan does not include Claude Code.
2. Interface chosen: desktop app for week one, terminal once you are comfortable.
3. Install verified with claude --version and claude doctor, or the Code tab opens in the app.
4. A practice folder with a copy of real data, not the original.
5. First session started in Plan or Manual mode, not Auto.
6. First three prompts are questions, not changes.
7. CLAUDE.md created with /init, kept under 200 lines (per the memory docs), with a “never touch” rule for raw data.
8. Every output checked by you: file opens, totals match, page loads.
9. /clear between unrelated tasks; a fresh prompt after two failed corrections.
10. Git or a manual copy in place before any change you cannot easily redo.
11. /usage checked weekly to see where your plan’s usage goes.
12. Claude Code 101 on Claude Academy completed for the parts you skipped.

FAQ

Do I need to know how to code to use Claude Code?

No. Anthropic’s terminal guide says directly that you “don’t need to know how to code” and can describe what you want in plain English. You do need to be able to judge whether the result is right, so start with tasks where you can check the output yourself.

Can I use Claude Code for free?

Not on the free claude.ai plan; Claude Code needs Pro or higher, or a Console account with API credits. Pro costs $20 per month, or $17 per month billed annually, according to Anthropic’s pricing page. The Claude Academy courses themselves are free to take.

How long does it take to learn Claude Code?

Most people can complete a useful first task in their first session. Anthropic’s Claude Code 101 course covers the fundamentals in 12 lessons and about 1.5 hours. Getting good at writing specific prompts and a solid CLAUDE.md takes a few weeks of regular use.

Is Claude Code safe to run on my work laptop?

It is as safe as the permission mode and folder you give it. Start it inside a project folder, not your home directory, use Manual or Plan mode while you learn, and avoid bypass mode outside isolated containers. Check your company’s policy too, since Team and Enterprise admins can enforce settings for everyone.

Does Claude Code work on Windows?

Yes. It runs natively on Windows 10 version 1809 or later from PowerShell or CMD, or inside WSL, per the system requirements. Installing Git for Windows is optional but gives Claude Code a Bash shell; the desktop app avoids the terminal entirely.

What is the difference between Claude Code and the Claude chat app?

Chat answers questions and drafts text you then copy elsewhere. Claude Code works directly on the files in a folder: it reads them, edits them, runs commands and shows you the changes for approval. Both are included in the same paid plan, and the desktop app has both in separate tabs.