Beginner walkthrough
Claude Code Tutorial for Beginners
Every step of installing and building with Claude Code, as illustrated screenshots with timestamped links back to the source video.
An illustrated walkthrough of 15 steps - about 5 minutes to read. Every step links back to the exact moment in the source video.
TL;DR
Install Claude Code with one command (or via npm), sign in with your Claude subscription, open it inside VS Code, and draft your first build in plan mode so Claude asks you questions before writing code. Save your project's rules to CLAUDE.md with /init, commit early so you can roll back, and only reach for bypass-permissions mode once you trust the loop. The whole flow below is screenshot by screenshot.
About the source video
This walkthrough follows the steps and screen flow of a beginner tutorial by Kevin Stratvert. Each step links back to the exact second in the video so you can watch the original.
Screenshots are frames from the video, credited to the creator; the write-up is our own. Steps verified against the video in September 2026.
Install Claude Code and sign in
One command in a terminal gets you the CLI; a browser sign-in connects your plan.
- 1
Open the install page and copy the script
From the Claude Code homepage, use the copy button to grab the install script and run it in a terminal. Windows users get a PowerShell variant - and if the script gives you trouble, the npm route below is the reliable fallback.
The install command lives right on the homepage, with Terminal and IDE tabs.Watch at 0:22 - 2
Make sure Node.js is installed
The npm install route needs Node. If you are not sure you have it, download the installer from nodejs.org, then confirm with node -v - the version number should print right back.
node -v prints your installed version; anything recent works.Watch at 1:00 - 3
Run claude and connect your account
Type claude and press enter. Pick a text theme, then choose how to sign in - a Claude subscription (Pro, Max, Team) or an Anthropic Console account billed by API usage. Once the browser hop completes, you land at the welcome prompt.
Subscription or Console: the welcome screen means you are connected.Watch at 1:22 - 4
Open Claude Code inside VS Code
The terminal works fine on its own, but the editor integration is where it shines: open a project in VS Code, create a new terminal, type claude, and press 1 to trust the directory. Claude now sees your open files and diffs.
Same CLI, but now it shares context with your editor.Watch at 2:30
Build a first app in plan mode
Let Claude interview you before it writes a line of code.
- 5
Write requirements as bullets, not a vibe
Vague prompts produce vague apps. List the actual requirements, add 'ask me questions to clarify the requirements, engineering principles, hard constraints', and hold Shift+Tab until the footer says plan mode. In plan mode Claude researches and asks instead of writing code.
Bulleted requirements plus the ask-me-questions trick, in plan mode.Watch at 3:28 - 6
Answer Claude's clarifying questions
Plan mode comes back with numbered questions - tech stack, storage, extras. Answer with the number keys, use the left and right arrows to move between questions, and press enter to submit. This interview is what turns a vague idea into a buildable spec.
React with Vite, local storage - the choices are one keypress each.Watch at 3:46 - 7
Review the implementation plan
Claude produces a plan: summary, key features, tech stack, data models, project structure, implementation steps. Read it for decisions you disagree with - the demo catches JavaScript where TypeScript was wanted. Choose the feedback option and Claude rewrites the plan.
Data models, file structure, phased steps - all editable before code exists.Watch at 4:42 - 8
Approve the plan and pick an edit mode
Option 1 auto-accepts every edit, option 2 asks per file. Manual approval is safer for your first run but tedious fast - auto-accept plus a git branch is the workable middle ground once you commit early and often.
Auto-accept edits or approve one by one - your call.Watch at 5:35 - 9
Approve commands as Claude runs them
By default Claude asks before running anything: installs, builds, dev servers. Option 2, 'don't ask again for npm install commands in this folder', is the sane default for package installs you trust. The video's author still prefers starting the dev server himself in a second terminal.
Per-command approvals, with a per-folder 'don't ask again' escape hatch.Watch at 6:05 - 10
Test the app it built
The demo build is an office chore manager: a calendar view, team members, recurring chores. Click through the real UI - add a chore, make it weekly, check the week view. This is where you find the small bugs worth fixing next.
A working calendar app, with a couple of bugs to hunt down next.Watch at 7:00
Onboard Claude with CLAUDE.md
One file, loaded every session, that teaches Claude your project.
- 11
Understand the three layers of CLAUDE.md
CLAUDE.md is injected into every Claude session, so keep it to universally applicable instructions. Three layers: the WHAT (tech stack, project structure, key files), the WHY (what each component is for), and the HOW (how you want Claude to work - for example, run tests after every change).
WHAT, WHY, HOW - the three layers every CLAUDE.md needs.Watch at 8:42 - 12
Generate it with /init (then improve it)
/init creates a CLAUDE.md from your codebase automatically. The video's honest take: on its own it misses the three-layer structure, so paste in a prompt that asks for the what/why/how plus progressive disclosure - an index that points to deeper markdown files.
/init gives you a starting CLAUDE.md in seconds.Watch at 9:20 - 13
Add house rules - like 'branch first'
Edit the generated file like any doc. The demo adds one line: when working on a new feature or bug, create a git branch first. From then on Claude branches proactively - exactly what you want before letting it auto-accept edits.
One added rule: always start features and fixes on a branch.Watch at 10:42
Go faster - safely
Bypass permissions for speed, git for the safety net.
- 14
Know what bypass permissions does
Launch with claude --dangerously-skip-permissions and Claude stops asking before running commands. The video frames it honestly: there is real risk, but for trusted loops the speed can be worth it. You can also cycle modes with Shift+Tab without the flag.
The red warning is real - know what you are approving.Watch at 12:28 - 15
Ship a real architecture change
The finale: make chores sync across tabs in real time. Claude explores, plans three options (polling, WebSocket, BroadcastChannel), then - with bypass mode on - builds the WebSocket server unattended. Two browsers side by side prove the sync works. That is the explore, plan, code, commit loop paying off.
Two browsers, one chore list, changes landing instantly in both.Watch at 14:06