Pages

Raw Markdown for agents: pages.md. MCP: read_guide("pages").

Pages are public forms, portals and HTML pages that live inside a Space and are shared through links. A page has a draft and immutable published versions. Flow: pages.create (draft) -> pages.update_draft (definition, name, settings; published links keep their frozen version) -> pages.publish (new version, the response lists the inputs and outputs) -> pages.links_create (share link with optional password, email gate, expiry, visit limit) -> hand the link to the user. pages.get shows draft, current version and counters; pages.html_get returns the HTML artifact of an HTML page.

Blocks pages show data with two block types: records (key, record_type_id, field_ids, optional view_key, base_filter, sort, page_size, allow_export) and chart (key, record_chart_id of a saved chart, optional days and height sm|md). Both are checked against the space on every save; pages.publish freezes the schema version and, for charts, the effective spec, so later chart edits never change what a published link shows.

HTML pages (kind 'html') are a self-contained document (max 5 MiB) rendered in a sandboxed frame: no network, no external scripts, styles or fonts, only inline code and data: URLs. The document talks to the page through window.adlass, every call returns a Promise: getContext() -> {title, locale, preview, viewer, fields, submission_mode, can_download, data: [{key, type}]}; records(key, {cursor, limit, q}) -> {fields, items, next_cursor}; chart(key, {days}) -> {columns, rows, truncated, warnings, window, display}; submit(values); upload(file, key); download(assetKey); track(event). records and chart only accept keys declared in definition.data (same block shapes as above); submit validates against definition.fields. Chart rows carry numbers as strings and dates as ISO days, columns carry role x|series|measure and labels; display is the chart's title, type and format, enough to draw it (see the records_charts guide). settings.chrome 'none' hides the header bar so the document owns the whole viewport. The author of an HTML page is trusted like a space manager: the bridge limits what the document can read, not what a link inside it does with it; publishing stays a human decision.

Drafts may be incomplete (empty keys, labels or title); completeness is checked by pages.publish, which returns field paths. Download blocks name a slot via asset_key and may stay unbound: visitors then see the download as not yet available. pages.assets_set (or the assets field of pages.update_draft) binds a document of the Space to a slot; after binding run pages.publish and pages.links_retarget, because existing links stay pinned to their version.

Links: pages.links_list, pages.links_get (url, managers only), pages.links_update (label, expiry, gates, status ACTIVE or DISABLED), pages.links_rotate (new token, old URL stops working), pages.links_retarget (another published version), pages.links_revoke (irreversible).

Results: pages.submissions_list and pages.submissions_get return visitor answers following the version's output schema; pages.submissions_set_status marks them processed. pages.visits_list and pages.stats report traffic. pages.archive disables all links (pages.publish restores); pages.delete is irreversible.