Choose a public site
Start with an HTTP or HTTPS URL. The analyzer validates the target, then looks for useful same-site pages through public navigation, links, and site-map signals.
How it works
From a website's visual DNA to a new build.
Site Stripper maps a public URL, observes page structure and visual details, and organizes the evidence into a comprehensive ZIP. Give that reference to Claude, Codex, OpenCode, Cursor, and any other AI agent to guide a fresh website in the same visual style.
Map pages and design details
Pages are examined one at a time for headings, visible content, layout, design tokens, controls, media references, responsive behavior, and available captures.
Review the reference pack
Bounded observations are rendered into page notes, cross-page guides, and machine-readable files. Coverage notes keep partial results and known limits visible.
Build the ZIP in your browser
The browser validates the generated file set and assembles the ZIP. Screenshot binaries are included only when the corresponding captures are available.
Turn reference details into a new website.
The ZIP gives an AI coding agent a clear starting point for the reference site's visual DNA. Page notes explain what appears and in what order. Design tokens, responsive observations, and captures add the detail needed to carry the same style and art direction into a brand-new site or web app.
Choose your agentUse Claude, Codex, OpenCode, Cursor, and any other AI agent that can read a project folder or ZIP.
Keep the source details closeStart with REBUILD.md, then give the agent page notes, token files, layout records, responsive notes, and available assets.
Review fidelity as you buildCompare the new site with included captures and notes. Results depend on site complexity, source coverage, and the agent's implementation.
Observed page details
Each note stays linked to its source page.
Headings, visible text, navigation, media references, and layout samples are recorded alongside the page where they were observed. That makes it easier to move from a package summary to specific source details.
- Page-level notes preserve content order.
- Structured records keep the same details available as data.
- Capture files appear when a screenshot completes successfully.
Page selection
Review which pages were included and why.
Each run records the candidates it considered and the pages it inspected.
From candidate pages to the selected set
Page candidates come from public site structure and are ranked for relevance. The selection report records which pages were included and the reason for each decision. This makes the scope of a run easy to review.
Discovery sourcesSite-map entries, primary navigation, and relevant same-site links help build the candidate list.
Page-specific notesWritten notes keep headings, content, and section order tied to the page they describe.
Coverage dataSelection and report files show what was analyzed and where a run reached its limits.
Hosted run limits
One hosted run can analyze up to 10 selected pages in up to 150 seconds. Browser time and screenshot capacity are bounded, so a slow or media-heavy source may return partial results. The report makes those limits visible instead of implying complete coverage.
One page at a time
The analysis uses a single browser session and processes pages sequentially. This keeps the run focused and allows the report to associate each observation and capture with its source page.
After observations return, the browser-side package builder renders the guides and structured records, validates the file set, and creates the downloadable archive.
What the observations mean
The pack describes what the analyzer could observe at capture time. It is evidence for a separate rebuild workflow, not an offline copy of the source website.
- Source-site forms are not submitted.
- CAPTCHA and bot-verification challenges are not bypassed.
- Arbitrary JavaScript interactions are not replayed.
- Screenshot files appear only when captures succeed and fit the run limits.
See the result
Explore the files a run can produce.
Open a real sample and compare the top-level guide, page-specific notes, structured data, theme output, and captured image.