# Tool reference

> Choose the smallest tool that answers the current question, or use the replication kit when you need the complete evidence set.

All tools return JSON text and MCP structured content. Tools that produce screenshots, frame strips, or visual diffs can also return MCP image content blocks.

## Understand the rendered system

### `analyze_site`

Analyze a live HTTP or HTTPS page. Returns the computed color palette, typography, CSS variables, spacing, radii, shadows, and reusable component families.

### `analyze_html`

Render supplied HTML and optional CSS without external network access. Useful for generated markup, offline experiments, and deterministic tests.

### `inspect_component`

Inspect elements selected by CSS, accessible role and name, or visible text. Returns matched cascade rules, custom properties, box geometry, platform fonts, pseudo-elements, descendants, and hover, focus, or active changes.

### `extract_layout`

Return a breadth-first layout tree with section labels, flex and grid details, scroll containers, sticky or fixed positioning, clipping, stacking contexts, and hidden or collapsed elements.

### `extract_responsive`

Compare layout at multiple viewport widths. Returns breakpoints, media-query summaries, and a layout tree per width, with optional screenshots written to disk.

## Capture materials and visible states

### `capture_screenshot`

Capture the viewport, the full page, or one element. The image can be returned to the client and optionally saved through `outputPath`.

### `extract_assets`

Download images, background images, inline SVGs, fonts, videos, and favicons. Writes an `assets/` directory and a manifest describing where every file appeared.

### `clone_section`

Produce a self-contained section with cleaned HTML, scoped CSS, dependent variables, keyframes and font faces, asset references, and a screenshot. With `outputDir`, URLs are rewritten to downloaded local assets.

### `capture_interactions`

Operate likely menus, dropdowns, modals, tabs, and accordions, then capture what they reveal. Useful for interface states that never appear in a static page load.

## Understand motion

### `extract_animations`

Create a static motion census: keyframes, animation users, numeric timings, transitions, Web Animations API entries, scroll timelines, detected libraries, and generated CSS.

### `capture_motion`

Record motion over real time after load, scroll, hover, or click. Returns sampled frames plus numeric transform, opacity, filter, and clip-path curves for tracked elements.

Use `extract_animations` to understand definitions. Use `capture_motion` to see the actual timeline.

## Build and verify replicas

### `create_replication_kit`

Capture the complete desktop and mobile evidence bundle in one load: screenshots, assets, design tokens, layout trees, animation inventory, and a `REPLICATE.md` implementation guide.

### `compare_replica`

Capture the original and replica deterministically, calculate an anti-aliasing-tolerant YIQ diff, and identify the worst regions with relevant style deltas. Use `selfCheck` first when the original page contains unstable visual content.

### `crawl_site`

Crawl a small same-origin page set and merge it into a site-wide design system. With `outputDir`, writes page-level evidence, screenshots, `site.json`, and a `SITE.md` guide.

## Tools that write files

The following tools can write to paths supplied by the caller:

- `capture_screenshot` with `outputPath`.
- `extract_assets` and `create_replication_kit`.
- `clone_section`, `extract_responsive`, `capture_interactions`, `capture_motion`, and `crawl_site` with `outputDir`.

Use absolute paths. Keep generated evidence outside application source directories unless you intentionally want it versioned.

## Next

- [Follow complete workflows](workflows.md)
- [Review security and limitations](security.md)