Clone any website with one command: the whole setup
The short version: a free template plus one command rebuilds any page you point it at as a real Next.js project on your own machine. It copies the design faithfully, down to the computed spacing of individual buttons. It does not copy a checkout, stock, a supplier or a customer, because a browser never receives those. Used on your own site it is an enormous head start. Used on somebody else’s brand it is a legal question nobody in the demos mentions.
The clip going around is real. You install a template, type one line, and a few minutes later a copy of the page is running on your laptop. We read the whole repository rather than taking the video’s word for it, and the mechanism is exactly what it claims to be.
What the demos leave out is the second half of the sentence: what comes out of it, and what does not. That is most of this memo, because it is the part that decides whether this saves you three months or three hours.
What it is
ai-website-cloner-template is a public GitHub repository under the MIT licence. It is a Next.js starter with one skill installed: a long, very specific instruction file that teaches an AI coding agent how to take a page apart and rebuild it.
The repository ships that same skill in eleven different formats, so it works in Claude Code, Codex, Cursor, Gemini CLI, Amazon Q, Cline, Continue, Roo, Windsurf, Kiro and GitHub Copilot. The README recommends Claude Code because the skill leans on driving a real browser and running agents in parallel.
The template is free and always will be: MIT licence, public repository, no paid tier. The AI that runs it is not. Claude Code requires a paid Claude plan, and the free claude.ai account does not include it. Better to know that now than after two downloads.
The setup, in four steps
This is the whole thing. There is nothing hidden behind a course or a Discord.
Make your own copy of the template
On the repository page, click Use this template, then Create a new repository. Give it a name and create it.
This step is not optional bureaucracy. The README asks specifically that you do not clone the template repository directly for your own website project, because your work then lives in somebody else’s repo instead of yours. Use the button and you get a clean project in your own account.
Open your new repository and install
git clone https://github.com/YOUR-USERNAME/YOUR-NEW-REPO.git cd YOUR-NEW-REPO npm install
If you have never used a terminal, the previous memo on vibe coding for complete beginners walks through installing an editor and opening a terminal inside it, step by step, with nothing assumed.
Start the agent with browser control
claude --chrome
The --chrome part is the important bit. It gives the agent a real browser it can open the target site in, which is how it reads the page’s true appearance rather than guessing from the source.
Point it at a page
/clone-website https://example.com
That is the entire instruction. You can pass several URLs at once and each one becomes its own route in the finished project.
What happens next, and why it is good
The skill describes its own method as being a foreman walking a job site. It does not read the whole page and then build the whole page. It walks the page section by section, and as it finishes measuring each one it writes a specification file and hands that file to a separate builder agent, which starts work immediately while the foreman moves on to the next section.
Three details are worth knowing, because they are why it works.
It measures rather than estimates. For every element it reads the real computed style out of the browser. The skill is blunt about this: not “it looks like 16px”, the actual computed value. It downloads the genuine font and image files rather than approximating them, and it captures behaviour too, so a nav bar that changes on scroll gets its trigger, its before and after states, and its transition recorded.
It splits the work by component, not by page. One agent per component, and a hard rule that any brief running longer than about 150 lines has to be broken into smaller pieces. The reasoning in the file is that an agent handed a whole section approximates spacing and guesses font sizes, while an agent handed one component with exact values gets it right every time.
It refuses to leave the build broken. Every builder has to pass a TypeScript check before it is finished, and after the work is merged the foreman runs a full production build. So what lands is a project that actually compiles, not a folder of approximations.
What you get, and what you do not
A browser only ever receives the front of the shop. That is not a limitation of this particular tool, it is the shape of the web: the server sends your browser some markup, some styles and some images, and keeps everything else. Anything pointed at a URL can copy the first list and can never touch the second.
So the packing list is short and worth being precise about.
The layout and structure. The computed colour, spacing, sizing and typography of every element. The real text. The actual image, icon and font files. Hover, scroll and animation behaviour. All of it as a Next.js project that builds.
The checkout. The stock. The supplier. The payment processor. The customer accounts and the order history. The traffic, the search ranking and the reason anybody buys from that site rather than another one. None of this is served to a browser, so none of it can be cloned by anything.
This is where the popular framing quietly falls apart. The pitch is usually some version of: no more three months spent thinking up a product, building it, making the web page and finding suppliers. Then, twenty seconds later, the same pitch says you just have to deploy it, create your products and put them online.
Read those two sentences next to each other. Of that four item list, cloning removes exactly one: the web page. The product, the suppliers and the launch are all still sitting there waiting for you. That is a genuinely useful thing to have removed, and it is one item, not four.
Then it is yours to edit
The output is an ordinary Next.js project in your own repository, so editing it is just editing code, and you can do that the same way you cloned it: by asking. New headline, new palette, your own images, your own copy. Nothing about the result is locked or obfuscated.
The one thing to be clear-eyed about is that “replace the products with yours” means adding a product system, not swapping entries in one you inherited. There is nothing to swap them into yet.
Whose brand is it
We searched the repository’s README, its skill file and its project instructions for any guidance on this, and there is none. The only licence mentioned is the MIT licence on the template itself, which covers the template and says nothing about the site you point it at. The most-shared demo of the tool clones YouTube.
So, plainly: a website’s design, copy, photographs and logo generally belong to whoever made them. Using this on your own site, on a page you have permission to rebuild, or as a private reference while you learn how a layout works, raises nothing. Publishing a pixel copy of a live business under your own name is a different act, and the fact that a tool makes it take four minutes does not change what it is.
We are not lawyers and this is not advice. It is simply the question that should have been in the demo.
What it is genuinely excellent for
Rebuilding your own site on a modern stack without redoing the design. Getting a real, working front end to prototype an idea against instead of a wireframe. Learning how a layout you admire is actually constructed, at the level of real numbers. Migrating a page you own off a platform you have outgrown.
In all four of those the thing you needed was the front end, and the front end is exactly what it delivers, faithfully and for free. That is a serious head start. It is just a head start on the web page, not on the business.
If you want the same trick pointed at your own workflow rather than somebody’s homepage, the memo on building an AI loop that checks its own work is the next thing worth reading.
Questions people actually ask
Is the website cloner actually free?
Yes. The template is a public GitHub repository under the MIT licence, and there is no paid tier for it. What is not free is the AI that runs it: Claude Code needs a paid Claude plan, and the free claude.ai account does not include it. So the tool costs nothing and the engine costs something.
What does the command actually copy?
Everything a browser receives. Layout, the computed colour, spacing and font of every element, the real text, and the actual image and font files, downloaded rather than guessed at. It rebuilds all of that as a Next.js project you own.
What does it not copy?
Anything the browser never sees. There is no checkout, no inventory, no payment processor, no customer accounts, no supplier, no order history, and no traffic. Those live on the other machine and are never sent to your browser, so nothing pointed at a URL can retrieve them.
Can I just swap in my own products and launch?
You can swap the words, the colours and the images, because what you get is an ordinary code project in your own repository. What you cannot do is swap products into a shop that does not exist yet. The clone gives you the storefront. Everything that makes it a shop is still yours to build.
Am I allowed to publish a clone of someone else's site?
That is a real question and neither the tool nor the videos about it raise it. The template's MIT licence covers the template, not the site you point it at. Website designs, copy, photographs and logos generally belong to whoever made them. Cloning as a reference, a template for your own project, or a rebuild of a site you already own is uncontroversial. Publishing a pixel copy of somebody's live brand is a different thing, and worth a lawyer's opinion rather than ours.
Does it only work with Claude Code?
No. The repository ships the same skill for Codex, Cursor, Gemini, Amazon Q, Cline, Continue, Roo, Windsurf, Kiro, OpenCode and GitHub Copilot. The README recommends Claude Code with Opus 5 because the skill leans on browser control and parallel subagents, which is where it was tuned.
How many agents does it run?
As many as the page needs. It is one agent per component, not a fixed number: a simple banner gets one, a section with three different card variants gets one per variant plus one for the wrapper. The skill has a hard rule that any brief longer than about 150 lines has to be broken into smaller pieces.
Want this built for you?
We write these memos because we build this stuff every day. If you want it working in your business instead of sitting on your reading list, that is literally our job.