Website Cloning Toolkit
Every verified skill and tool, mapped. Enter hub password to continue.
That's not right. Try again.
How to Clone a Website: The Complete Map
First: which job do you actually mean?
Start here"Clone a website" is really three different jobs, and the right tool depends entirely on which one you want:
- A. Rebuild the design into your own editable code (recreate the look, adapt to your brand). This is the design-to-code / design-extraction job.
- B. Mirror the whole site as-is for offline reference, archive, or rehosting. This is the download / mirror job.
- C. AI-generate the code from a screenshot or URL. This overlaps with A but uses AI regenerators that invent new code that resembles the source.
Below, each job gets its own verified toolset, then a recommendation tuned to your stack (Astro + Cloudflare Pages) and your preference for free, self-hostable tools.
The one honest truth up front: animation rarely survives
Read thisAcross every tool the sweep verified, the single most important distinction is this:
Regenerate vs copy-the-source
AI regenerators (v0, Anima, screenshot-to-code, the Claude cloner templates) look at the site and write new code that resembles it. They invent plausible layout and often invent new animations. Result: "looks similar," not "is the same." Real motion, scroll effects, and JS behavior do not carry over.
Source copiers (Firecrawl, browser Save-As, wget) pull the actual HTML/CSS/JS. This is the only way real animation and interactivity survive, because you keep the real code. The tradeoff: you inherit the source's messy production code, not clean components.
The one AI tool that even attempts motion is abi/screenshot-to-code, which accepts a screen recording (not just a still) and tries to rebuild the interaction. Still approximate. If faithful animation is the goal, copy the source; don't regenerate.
Job A1: Clone from inside Claude Code (skills you can install)
Design → codeThese run in Claude Code using your own subscription (free to run). All star counts verified live July 9, 2026.
| Skill | Stars | What it does | Clones | Animation |
|---|---|---|---|---|
| ai-website-cloner-template | 27,000+★ | One command /clone-website <url> drives Chrome to read the real DOM + computed styles, then parallel builder agents rebuild it with a visual-diff check. The most serious full cloner in the ecosystem. | Full site → Next.js/React/Tailwind | Flattens |
| claude-skill-web-clone | 590★ | "Real source first" method: reuses the actual site markup instead of hallucinating code. Only one explicitly claiming React + WebGL handling. | Full site, code-fidelity | Better |
| taste-skill | 218★ | /taste <url> outputs a design profile (.md + .json) capturing tokens AND the design rationale (the why behind them). | Design language only | No |
| anydesign | 128★ | Reads images, URLs, or Figma, emits a structured design.md with tokens + component inventory. Most flexible on input source. | Design → spec | No |
| ui-cloner installed | 1★ | Chrome-based "clone any site and adapt to your brand." You just installed and tested this one. Inert markdown, unvetted author, but free to run. | Design + layout + attempts animation | Attempts |
| extract-design-system installed | 119★ | URL to W3C design tokens (tokens.json + tokens.css). Clean, proven, actively maintained. | Tokens only | No |
Catalogs to keep harvesting: davila7 templates, awesome-claude-code-and-skills. Flagged unverifiable: "Manavarya09/design-extract" (heavy SEO presence but 404s on the GitHub API right now, do not install).
Job A2: AI design-to-code tools (outside Claude)
URL/screenshot → codeOpen-source / self-hostable (your preferred lane):
| Tool | Price | Input | Output | Note |
|---|---|---|---|---|
| Firecrawl installed | OSS / free | URL | Real HTML + markdown + JSON + screenshot | The faithful path: gives you the true source, so real animation survives. Code is production spaghetti, not tidy. You already have the skills + MCP. |
| abi/screenshot-to-code | OSS / self-host | Screenshot, Figma, screen recording | HTML/Tailwind, React, Vue, Bootstrap | 73k★. Best OSS regenerator; the only one that attempts motion via screen-recording. Bring your own model key. |
| Dembrandt | OSS npm / free | URL | DTCG tokens, Tailwind config, DESIGN.md, MCP | npm i -g dembrandt. Best free brand-token extractor (tested + security-clean), works natively with Claude Code. Caveat from our run: over-promotes neutrals (labeled gray as "primary") and mis-guessed the framework. Trust its tokens, not its role labels. |
| Project Wallace / CSS Stats | Free | URL / CSS | CSS analytics + token audit | Analysis only, not a rebuild. Good precursor to hand tokens to a builder. |
Hosted SaaS (best-looking output, but reinterprets rather than copies):
| Tool | Price | Input | Output | Note |
|---|---|---|---|---|
| v0 (Vercel) | Free tier + paid | Prompt, screenshot, URL | React + Tailwind + shadcn | Best-looking, most production-ready output. Invents its own animations. Not self-hostable. |
| Anima | Freemium | Figma, image, URL | React, HTML, Vue, Tailwind | One of few with an explicit URL "clone website" button. Good layout, invents interactivity. |
| Builder.io Visual Copilot | Freemium | Figma | React, Vue, Svelte, Angular, HTML | Cleanest, most codebase-ready output. Figma-first, so pair with html.to.design to turn a URL into Figma first. |
| html.to.design | Freemium plugin | URL | Editable Figma layers | The bridge: URL → Figma → any Figma-to-code tool. Static snapshot. |
Job B: Mirror the whole site (offline / archive / rehost)
DownloadAll free and Windows-friendly unless noted. This is about downloading the real pages + assets, not rebuilding.
| Tool | Best for | Access | JS/SPA? |
|---|---|---|---|
wget --mirror | Static sites, zero install | Already on your machine (git bash) | No JS |
| HTTrack | Static sites, friendly GUI | winget install HTTrack.WinHTTrack | No JS |
| SingleFile CLI | Faithful single JS-rendered page | npm i -g single-file-cli | Yes (real browser) |
| Firecrawl installed | Whole JS site → clean HTML/markdown, from inside Claude | firecrawl:firecrawl-download / -crawl | Yes (renders JS) |
| Browsertrix Crawler | True interactive replay (WACZ) | Docker Desktop | Yes (highest fidelity) |
| monolith / obelisk | One self-contained HTML file | winget install Y2Z.Monolith | No JS alone |
What to actually reach for (tuned to your stack)
RecommendationThe free, autonomous trio you already own
1. Firecrawl (installed) to pull the site's real HTML/CSS/JS + a screenshot.
2. Dembrandt (npm i -g dembrandt, free) to lock the exact brand tokens.
3. Claude Code (your ui-cloner or extract-design-system skill) to rebuild it into your own Astro components.
100% free, runs on your subscription, and produces cleaner-than-copy results you can actually adapt.
If you want a turnkey full clone and don't mind budget + porting: JCodesMore's ai-website-cloner-template (27k★) is the best in class, but it outputs Next.js (you're on Astro, so it needs porting) and costs roughly $5 to $30 per clone in model tokens.
If you just need the design language to inform your own build (the safest and honestly highest-quality play): taste-skill or Dembrandt to extract tokens + rationale, then build fresh in your voice.
The copyright reality (plain, not a lecture)
Worth knowingCloning a third party's site wholesale copies their copyrighted content and usually breaks their Terms of Service. Trade-dress and copyright claims are real if you republish someone else's look-and-feel.
Low risk: mirroring for private/offline reference, and extracting tokens + layout patterns to inform your own original build.
High risk: pixel-copying a competitor's site and rehosting it publicly. Respect robots.txt, don't hammer servers, and when in doubt, extract-and-adapt rather than copy.
Compiled July 9, 2026 · Private · All repos verified live via GitHub API · ← Hub