Claude Code is Anthropic’s AI coding agent that runs on your laptop and ships software when you describe what you want in English — and this guide takes you from never having opened a terminal to your first working Claude Code script in 4 hours, with no prior coding required. I’m Maxime, a 10-year founder, 4 years deep in AI, who learned Claude Code from this exact starting point in late 2024. By month 6, Claude Code was my most-used tool. By month 12, I’d shipped 47 working projects through it (see vibe coding workflow real).
This is the absolute beginner’s guide. If you’ve read Claude Code first 30 days, Plan Mode tutorial, or Cursor for non-engineers, this one sits BEFORE all of them. If you’ve never opened Terminal on your Mac or Command Prompt on your PC, start here.
What you’ll need before starting
- A laptop (Mac, Windows, or Linux). Mac is easiest for this guide; Windows needs WSL setup (covered below).
- An Anthropic account (we’ll create one in hour 1).
- ~4 hours of focused time. Doesn’t have to be consecutive — works as 4 × 1-hour sessions across a week.
- $20 for the Claude Code Pro subscription. Wait until hour 1 to subscribe; don’t subscribe before starting.
You do NOT need:
- Any coding background
- A “developer setup” (we’ll handle install ourselves)
- A specific project in mind (we’ll pick one in hour 2)
- A GitHub account (helpful later but not required for this guide)
Hour 1 — Install and orient
Step 1 — Open your terminal (5 minutes)
A “terminal” (also called “Terminal,” “shell,” “command line,” or “console”) is a window where you type commands and the computer responds. It’s the original way humans talked to computers before windows and icons.
On Mac: Press Cmd+Space (Spotlight), type “Terminal,” press Enter. A dark window opens with a blinking cursor.
On Windows: You need WSL (Windows Subsystem for Linux). Press Windows key, type “PowerShell,” right-click, “Run as administrator.” Then type: wsl --install and press Enter. Wait 5-10 minutes for it to install. Restart your computer. After restart, open “Ubuntu” from the start menu. This is your terminal.
On Linux: Press Ctrl+Alt+T (most distributions).
You should see something like:
yourname@MacBook-Pro ~ %Or similar. This is the prompt. You type commands here.
Step 2 — The 5 commands you need to know (10 minutes)
Don’t try to learn 50 terminal commands. Learn these 5:
| Command | What it does | Example |
|---|---|---|
pwd | Print working directory — shows where you are | pwd |
ls | List files in the current directory | ls |
cd | Change directory | cd Desktop |
mkdir | Make a new directory (folder) | mkdir my-projects |
claude | Start Claude Code (after install) | claude |
Try them now:
pwd
ls
cd Desktop
ls
mkdir claude-projects
cd claude-projects
pwdIf those worked, you can move on. The terminal is no longer scary.
Step 3 — Install Node.js (15 minutes)
Claude Code is a Node.js tool, so you need Node.js installed.
On Mac: Visit https://nodejs.org, download the LTS version (the “left button”), open the .pkg file, follow the installer.
On Windows (WSL): In Ubuntu/WSL terminal, type:
curl -fsSL https://deb.nodesource.com/setup_lts.x | sudo -E bash -
sudo apt-get install -y nodejsOn Linux: Use your package manager or the same curl command above.
Verify installation by typing:
node --versionYou should see something like v20.x.x or v22.x.x. If you get an error, restart your terminal and try again.
Step 4 — Sign up for Anthropic (5 minutes)
Visit https://claude.ai. Sign up (Google or email). Verify email.
Then visit https://console.anthropic.com → Settings → Plans → upgrade to Claude Pro at $20/month. (Not Max 5x yet — start with Pro.)
Important: this Pro subscription is for Claude.ai web AND Claude Code together. One subscription covers both.
Step 5 — Install Claude Code (5 minutes)
In your terminal:
npm install -g @anthropic-ai/claude-codeWait for it to finish (30-60 seconds). Then:
claude /loginThis opens your browser. Sign in with the Anthropic account from step 4. Confirm. Return to terminal.
Then test:
claudeClaude Code starts. You should see a welcome message and a prompt. Type “Hello” and press Enter. Claude responds.
If you got this far, the install worked. Take a break.
Hour 2 — Your first working script
Step 1 — Pick your first project (5 minutes)
Don’t pick “build a SaaS.” Pick something tiny. The 5 candidates I’d recommend for a beginner’s first project:
| Project | Time | Why |
|---|---|---|
| A script that reads a text file and counts the words | ~20 min | Teaches file I/O |
| A script that fetches the current weather for your city | ~25 min | Teaches API calls |
| A script that converts a CSV file to a Markdown table | ~25 min | Teaches data transformation |
| A script that opens 5 of your favorite websites in tabs | ~15 min | Teaches shell automation |
| A script that renames all photos in a folder by date | ~30 min | Teaches file system work |
Pick one. I’ll walk through the first (word counter) for this guide.
Step 2 — Start a Claude Code session (2 minutes)
In your terminal:
cd ~/Desktop/claude-projects
mkdir word-counter
cd word-counter
claudeYou’re now in a Claude Code session in the word-counter directory. Anything Claude Code creates will go in this folder.
Step 3 — Ask Claude Code to build it (10 minutes)
Type this:
“I want a Python script called
count_words.pythat takes a text file path as a command-line argument, reads the file, counts the total number of words, and prints the count. If no argument is given, print a friendly error. Add a comment at the top explaining what the script does in plain English. Use Python 3 standard library only — no external packages.”
Claude Code will think for a few seconds, then propose to create the file. Read the proposed code. If it looks roughly like what you asked for, approve.
Step 4 — Test the script (5 minutes)
Claude Code might ask you to create a test text file. If not, create one yourself:
echo "Hello world this is a test file" > test.txtThen run the script:
python3 count_words.py test.txtYou should see output like:
Word count: 7If it worked, you’ve shipped your first Claude Code script. Celebrate. Take a break.
Step 5 — What just happened (5 minutes)
You described what you wanted in English. Claude Code wrote the Python code. You ran the code. It worked. The total time from “I want a word counter” to “I have a working word counter” was about 20 minutes.
This is the basic pattern. Every Claude Code task follows this shape:
- State what you want (specific, with file paths and acceptable inputs/outputs)
- Review what Claude proposes (read the code, understand the shape)
- Approve or edit
- Run and verify
The skill that compounds isn’t writing code. It’s writing tight specifications.
Hour 3 — Write your first CLAUDE.md
The CLAUDE.md file is the most important file in any Claude Code project. It tells Claude Code who you are, what your project is, and what you don’t want it to do. Skipping it is the #1 reason beginners quit.
Step 1 — Create the file (5 minutes)
In the word-counter directory:
touch CLAUDE.mdOpen the file in any text editor. On Mac:
open -e CLAUDE.mdOn Linux/WSL:
nano CLAUDE.mdStep 2 — Write a starter CLAUDE.md (30 minutes)
Copy this into the file and edit the bracketed parts:
# CLAUDE.md — Word Counter Project
## What this is
This is a learning project for [your name], a non-engineer founder learning
Claude Code. The goal is to ship small, useful Python scripts that automate
tasks I currently do by hand.
## The stack
- Python 3 (whatever's installed locally)
- Standard library only — no external packages without asking first
- macOS / Linux compatible
## Non-negotiable rules
- Never delete files without asking
- Never run `rm -rf` or any destructive shell command
- Never install global packages without asking first
- Always explain what each change does in plain English before writing
- If I ask for something unclear, ask a clarifying question instead of guessing
## My preferences
- Prefer simple, readable code over clever code
- Comment every function with what it does in plain English (3 lines max)
- Use clear variable names — `word_count` not `wc`
- Error messages should be readable by humans, not stack traces
## Definition of done
- Code runs locally without errors
- I can describe what the script does in one sentence
- There's a comment at the top of every file explaining what it's forThat’s the minimum. Save the file.
Step 3 — Test that Claude Code reads it (5 minutes)
Restart Claude Code:
claudeAsk:
“What’s in my CLAUDE.md file?”
Claude Code should summarize the file. If it does, the file is being read. Every future session will reference it.
Step 4 — Iterate on your CLAUDE.md (15 minutes)
The CLAUDE.md file grows with you. As you encounter outputs you don’t like, add rules. Examples that I added to mine over time:
- “Always add type hints in Python function signatures”
- “Never use single-letter variable names except for loop counters (i, j, k)”
- “Comment any line that uses a non-obvious regex or library call”
- “Date format in any output: ISO 8601 (YYYY-MM-DD)”
The file is a living document. By month 3, yours will be 100-200 lines.
Hour 4 — Ship a second project that’s actually useful
The first project (word counter) was a toy. Project #2 should be something you’ll actually use this week.
Pick from these (15 min decision):
- Daily startup script: Opens your 5 most-used apps and tabs in the morning
- Inbox summarizer: Reads your unread emails, summarizes top 5 in one sentence each
- Calendar tidier: Reads your Google Calendar, exports next-7-days as a markdown file
- Notion to Markdown: Exports a specific Notion page to local Markdown for backup
- Daily journal prompt: Generates a journal prompt based on today’s date and recent calendar events
Each of these is a real solo founder workflow tool. They’re not glamorous. They save 15-45 minutes per day when running.
Build one (90 minutes):
Same loop as Hour 2:
- State the task with file paths, inputs, outputs
- Review the code Claude proposes
- Test the code
- Iterate until it works
If something breaks, paste the error message into Claude Code and ask for a fix. Don’t get stuck for more than 15 minutes; ask for help.
Test and ship (15 minutes):
Run the script. Confirm it does what you wanted. Use it for one real task today. If it saved you time, you’ve shipped your first useful Claude Code project.
The day-30 checklist
By the end of your first month with Claude Code, all of these should be true:
| Item | Target |
|---|---|
| You’ve used Claude Code on 10+ different days | Yes |
| You’ve shipped 3+ projects you actually use | Yes |
| Your CLAUDE.md is at least 50 lines | Yes |
| You can describe each project in one sentence | Yes |
| You haven’t accidentally deleted anything | Yes (you followed the rules) |
You’ve never typed rm -rf | Yes |
| You’ve read every diff Claude Code proposed before approving | Yes |
If you’re at 7/7 on day 30, you’ve graduated from beginner to intermediate. The next reading is Claude Code first 30 days (the intermediate version) and Plan Mode tutorial.
Troubleshooting (the common blocks)
Problem: claude: command not found after install.
Fix: Restart your terminal. If still broken, run npm install -g @anthropic-ai/claude-code again and check it completes without errors.
Problem: Claude Code asks for permission on every action.
Fix: This is correct behavior at the beginning. Don’t click “always allow” on shell commands for the first 30 days. The friction is your safety net.
Problem: Claude Code says “I can’t access that file.”
Fix: You’re probably in the wrong directory. Run pwd to see where you are. Run ls to see what files exist. Run cd /path/to/project to navigate to the right place.
Problem: Script runs but produces wrong output.
Fix: Tell Claude Code exactly what happened. Don’t say “it didn’t work.” Say “I ran X and expected Y, but got Z. Here’s the actual output: [paste output].” Specific reports get specific fixes.
Problem: “I’m hitting rate limits.”
Fix: Pro plan limits reset every 5 hours. If you’re running into limits in week one, you’re probably running too many long sessions. Take a break or upgrade to Max 5x (but only after day 30).
Problem: “Claude Code seems to forget what I said earlier.”
Fix: Sessions have memory; new sessions don’t. Either continue your current session or move durable context into CLAUDE.md.
The single-paragraph beginner verdict
Claude Code is the AI coding agent that turns “I want to do X” into “X is done” without you typing code. From zero to first working script: ~2 hours. From zero to “I use this every day”: ~30 days. Cost: $20/mo Pro plan. The skill that compounds is specification writing, not coding. The CLAUDE.md file is the most important file you’ll write. The 5 terminal commands that matter: pwd, ls, cd, mkdir, claude. After 30 days, graduate to Claude Code first 30 days and Plan Mode tutorial. The compounding is real.
For the wider stack, see my live tool stack, vibe coding workflow real, and Cursor for non-engineers tutorial.
FAQ
Do I need to know what a terminal is before starting?
No. By the end of this guide you'll know what a terminal is, how to open it, and the 5 commands that matter. The guide assumes you've used a computer but never written code. If you can use Gmail, you can finish this guide.
Will I break my computer?
No. Claude Code asks before running any destructive command. The guide explicitly tells you to never click 'always allow' on shell commands for the first 30 days. With that rule in place, the worst that happens is a wasted hour.
How long until I can actually do useful things?
Hour 2 — your first working script. Hour 4 — you've shipped something you'll actually use again. Day 7 — Claude Code feels natural. Day 30 — it replaces a dozen manual tasks per week. The 4-hour mark is the 'this is useful' moment.
What if I get stuck?
Three resources, in order. One: this article's troubleshooting section at the bottom. Two: Claude itself — ask Claude (in browser) for help. Three: the Anthropic Discord community. Most beginner blocks have an obvious fix you can find within 10 minutes.
Should I use Pro or Max 5x as a beginner?
Start with Pro at $20/mo. For the first 30-60 days, you won't hit the Pro plan's usage limits. Upgrade to Max 5x at $100/mo when you've shipped 5+ projects and find yourself hitting limits. Most beginners burn $20 and quit before realizing the tool actually works — start with the lower tier.