sitewright

Show whether this agent is connected and, if so, the project, role, and capabilities. Call this first.

Connect this agent to a Sitewright project. Returns a URL + code for the user to approve in their browser.

Re-authenticate to connect to a DIFFERENT project (project scope is fixed per connection). Returns a URL + code to approve.

The authoring contracts of the first-party interactive components (carousel, tabs, lightbox, modal, banner, form): markers, data-sw-part roles, config attributes, and copy-paste markup skeletons. Optionally filter by type or marker.

Find icon names for {{sw-icon "name"}} (Phosphor, filled by default; add a ":weight" suffix like ":bold" or ":duotone"). Pass ONE OR MORE search terms, comma- or whitespace-separated (e.g. "settings, trash cart") — returns matching names per term. Searches Phosphor names, Lucide-name aliases and key…

Find transparent background TEXTURES (tileable PNG overlays — paper, fabric, noise, denim, grid…) to set as an element background. Pass ONE OR MORE terms, comma- or whitespace-separated (e.g. "paper, fabric denim") — returns matching texture names per term, each with ready-to-paste CSS. The colour c…

One INDEX of everything this platform can do and WHERE each is documented: the interactive components, the get_guide topics, the {{sw-*}} reference, how each content kind is written, and a need→tool lookup (e.g. "ripple" → get_guide effects, "collections" → get_guide datasets). Call this before assu…

The authoring REFERENCE for writing a page `source`: the {{sw-*}} HELPERS, the data-sw-* editable DIRECTIVES, the BINDING namespaces (company / website / page / page.data / pages / dataset / item / nav …), and the {{#each}} LOOP VARIABLES. Derived from the live engine, so it always matches what ship…

List the project’s pages. Returns METADATA only by default (id/path/title/status/nav/parent/order/template/…): a page’s Handlebars `source` and `data` store are omitted and described under `_summary` instead, because a full listing of a real site runs to hundreds of KB and blows the tool-output limi…

Get one page by id. For code-first pages the design is in the `source` field. The response also carries `previewUrl` — a signed DRAFT preview of this page that needs no login and works with no deploy target.

List all entities of a content kind. For kind 'entry' pass `dataset` (a dataset slug) to list ONLY that dataset's entries — an entry id is unique only within its dataset, so an unscoped entry list returns EVERY dataset's rows mixed together. Pass summary:true to omit the heavy body fields (source / …

Get one content entity by kind + id. For an ENTRY also pass `dataset` (its owning dataset slug) — an entry id is only unique WITHIN its dataset, so two datasets can share an id.

List a content entity's revision history, newest first (id, op, who, when, note). Pair with restore_revision to roll back a bad edit. For an ENTRY also pass `dataset` (its owning dataset slug).

Restore a content entity to an earlier revision (its id from list_revisions). Non-destructive: the current version stays in history, and a deleted entity is recreated. For an ENTRY also pass `dataset` (its owning slug) — the same one used with list_revisions.

The OBJECTIVE clone-fidelity gate for an imported page: renders your BUILD and the ORIGINAL source, measures computed styles per element + whole-bar chrome facts, and returns a measured PASS/FAIL — body (font/gradient/coverage) and chrome (position/size/style + skew, font-weight, letter-spacing, rad…

The OBJECTIVE clone-acceptance prerequisite — the facts a screenshot can't show and computed-style coverage can't game. fidelity_check only measures computed styles, so it passes a clone whose datasets are duplicated, whose modals were dropped, whose slider is dead, whose fonts don't actually load, …

Lighthouse PAGE-SPEED + SEO audit of a page, run against a DEPLOY-EQUIVALENT build (minified like Publish, with production cache headers — not the sandboxed draft preview). Returns four category scores 0–100 (performance, accessibility, best-practices, seo), core lab metrics (FCP / LCP / TBT / CLS /…

The VISUAL acceptance gate for a cloned/imported page — the reliable fidelity signal the computed-style scorers (fidelity_check / clone_audit's visual leg) miss (they score fonts/gradients/coverage of TEXT elements and are BLIND to layout, images, section design, and modals; a hollow page scores gre…

MEASURE a rendered page — settled markup + REAL computed styles + REAL rects for the CSS selectors you name. This is how you get NUMBERS off the LIVE ORIGINAL (font-size, padding, gap, colour, gradient stops, border-radius, shadow, transform), which the other fidelity tools cannot give you: they all…

HIGH-RESOLUTION visual compare of an imported page's chrome: crops the nav HEADER and FOOTER (or pass regions to limit) of your BUILD and the ORIGINAL, at 2× device scale as lossless WebP, and returns them build-then-original per region. Use it to SEE fine detail that compare_to_source's 1× full-pag…

Read where this project actually stands. `status` is the headline: "unpublished" whenever no deploy target is configured — publishing BUILDS the site, but nothing serves it until a target exists, so there is no live address. `url` is non-null ONLY for Local Hosting; it is null for a remote (FTP/SFTP…

List form submissions for the project, newest first. Optionally filter by formId and paginate with limit/offset.

List the configured stock-image providers and whether each is available (openverse needs no key; unsplash/pexels/pixabay need an instance-admin key). Only needed to explain a gap — search_stock_images with provider "all" already uses every available one.

Search stock photos. `provider: "all"` (the default choice) queries every available provider at once and interleaves the results; name one provider to search only that. Each hit carries its own `provider` — pass it back to import_stock_image along with the id. Returns provider-hosted `thumbUrl` (gri…

List the project’s self-hosted media assets — each with the URL to reference in an <img src> / href, plus kind, dimensions and alt. Optionally filter by kind = image | file | font.

List the project’s media FOLDERS (virtual grouping labels — slash-delimited paths, "" = root). Call before organizing assets so you reuse existing folders instead of creating duplicates.

Create or REPLACE a page. The page id is taken from page.id. This is a TOTAL replace — every field you omit is deleted, so only use it when you are writing the whole page. TWO fields survive an omission: `parent` (a page always hangs off a home, so a replace never yanks a sub-page up to the root and…

PATCH an existing page: send only the fields you want to change and everything else is kept. Use this instead of put_page for partial edits — put_page REPLACES, so `{id, path, title, nav}` would silently wipe `source`, `status`, `description` and `order` (`parent` and the `data.swImport` import mark…

Delete a page by id. Needs the content:delete capability.

Create or replace a content entity of the given kind. Args: { kind, id, data } — plus `dataset` (the owning dataset slug) when kind is "entry". For PAGES prefer put_page (fully typed). `data` must match that kind’s schema; you may OMIT `data.id` (and an entry’s `data.dataset`) — they are copied from…

Delete a content entity by kind + id. For an ENTRY also pass `dataset` (its owning dataset slug). Needs the content:delete capability.

Delete MANY entities of one kind in ONE call: { kind, ids:[…] } — plus `dataset` (the owning dataset slug) when kind is "entry". Use this instead of looping delete_content when clearing up after an import (junk datasets, a batch of entries or scaffolded pages): one call instead of N, so you do not b…

Add a translation-target LANGUAGE to the site — the ONLY correct way to do so. In ONE atomic step it registers the locale AND scaffolds an inherited translated page for EVERY existing page (the /<locale>/… subtree; each variant inherits the main language's code, so you then only fill in its translat…

Remove a translation-target language: drops the locale from settings AND cascade-deletes every page in that language's /<locale>/… subtree (and prunes its translation-catalog column). The default (main) language cannot be removed. Needs the content:delete capability.

Import a stock photo (by provider + id from search_stock_images) into the project. The server downloads, optimizes, and self-hosts it as a media asset with attribution — never a hotlink.

Crawl + IMPORT a public https website URL into THIS project — the FIRST step of cloning/nativizing a site. The server fetches and RENDERS the live page(s) itself (executing JS, following an embed/preview wrapper to the real framed site), self-hosts the images + fonts, and creates the imported `swImp…

Import an image into the project from a PUBLIC https URL — the server downloads, optimizes, and self-hosts it (never a hotlink), returning the stored asset (use its `url` in your <img src>). For STOCK photos use search_stock_images + import_stock_image instead. This is the ONLY URL-based import, and…

Upload a LOCAL file (a logo, a photo, a font, a PDF) into the project media library. Use this when the file is on YOUR disk; use import_image when it is at a public https URL. TWO STEPS: (1) call this to get a one-shot `uploadUrl`, (2) send the file to it yourself, e.g. `curl -T ./logo.png "<uploadU…

REPLACE a LARGE local file in place (the ticket lane for replace_media — a photo, a video, a PDF). TWO STEPS: (1) call this with the asset `id` to get a one-shot `uploadUrl`, (2) send the file to it yourself, e.g. `curl -T ./hero.jpg "<uploadUrl>?filename=hero.jpg"` — always pass ?filename=, since a…

Create an (empty) media folder + any missing ancestors. `path` is a slash-delimited grouping label (segments: letters, digits, space, _ or -), e.g. "About/Gallery". Folders are virtual labels; the file bytes stay where they are.

Rename OR move a media folder: re-roots the folder subtree AND re-files every asset under it. `from`/`to` are full folder paths. Fails if `to` already exists.

Move and/or rename a single media asset: `folder` re-files it (use list_media_folders / create_media_folder), `filename` sets its display name. Pass at least one. The asset URL is unchanged.

Rotate (90/180/270 clockwise) and/or crop an image asset. Rotation is applied FIRST, so `crop` is measured against the image AS TURNED — the same order an editor shows you. By default it edits IN PLACE: the asset id, stored name and URL do not change, so every page/dataset reference keeps working an…

Delete a single media asset — it moves to the File Manager Recycle Bin (RECOVERABLE for 90 days, then auto-purged). It is hidden from the media list and EXCLUDED from the next publish (a still-referenced page would then show a broken image on the republished site), so make sure NO page/dataset still…

Rename a dataset's slug AND/OR its display name. The slug must be an UNDERSCORE identifier (e.g. `faq_passengers`, NOT `faq-passengers` — it is used as a `dataset.<slug>` Handlebars path). This CASCADES automatically: every entry's `dataset` field and every page/template source's `{{#each dataset.<s…

Build the project’s static site from current saved content.

Autonomously clone/nativize EVERY imported page to the acceptance gate (server-side import→author→gate→iterate→publish). Requires a configured AI provider; a generic MCP agent can instead run the clone_site workflow itself.