AccessioAI

List accessibility scan sessions

List accessibility scan sessions for the authenticated organization, newest first. Use this when the user asks to see recent scans, find a specific scan to investigate, audit which pages have been scanned and when, or paginate through scan history. Paginate with `cursor` (last id from the previous page). Read-only.

Get an accessibility scan session by id

Fetch a single accessibility scan session by id. Use this after `list_scans` when the user wants details on one specific scan — issue counts by severity, scan duration, the page URL, the embed script that triggered the scan. Returns not_found when the id is unknown OR belongs to a different organization (the two cases are deliberately indistinguishable to avoid leaking cross-tenant existence). Read-only.

List AI-generated alt-text suggestions

List AI-generated alt-text suggestions, newest first. Use this when the user wants to triage the review backlog (filter `status=PENDING`), audit recent decisions (`status=APPROVED`/`EDITED`/`REJECTED`), or inspect alt-text for a specific embed script (`embedScriptId`). Pair with `approve_alt_text` or `reject_alt_text` to act on individual rows. Read-only.

Get the organization's current-month usage + plan limits

Returns the organization's current-month usage snapshot: currentMonth, currentUsage, quotaLimit, remaining, tier, isOverQuota. Use this BEFORE calling billable tools (`run_scan`, `approve_alt_text`) to confirm headroom, OR when the user asks about plan usage / how much quota is left this month. Read-only.

List embed scripts for the organization

List the organization's installed accessibility widget embed scripts. Use this when the user asks which sites have the widget installed, wants to look up an embedScriptId to filter `list_alt_text` or `list_scans`, or needs to see widget trial/expiry/status. Each row is an embed script row — website URL, status (ACTIVE/TRIAL/EXPIRED/SUSPENDED), trial/expiry timestamps, whitelabel branding. Does NOT return the embed token or its hash — those are bearer secrets that stay on the widget loader. Read-

List AI content declarations and detections

List AI content items for the organization, newest first. Use this to triage declarations/detections, audit status, or inspect items by obligation. Filter with `status` (PENDING, CONFIRMED, LABELED, EXEMPT, DISMISSED), `obligation`, `basis`, or `embedScriptId`. Pair with `confirm_ai_content`, `exempt_ai_content`, or `declare_ai_content` to act on the backlog. Requires `ai-act:read`.

Export AI Act transparency report

Returns a regulator-ready AI Act transparency report for the organization: all AI content items, label/exemption status, summary counts, and a tamper-evident evidence ledger snapshot with chain-verification status. Use this when the user asks for compliance evidence or an AI Act audit trail. Requires `ai-act:read`.

Queue a re-run of an existing scan session

Queue a re-run of an existing scan session. Use this when the user wants fresh accessibility results for a page they've already scanned (e.g. after deploying a fix), or when results are stale. Fire-and-forget — returns the scanner's acceptance envelope (`status`, `acceptedAt`); the actual results land in the existing scan session and surface on subsequent `list_scans`/`get_scan` calls. Spends AI credits — check `get_quota` first if quota is a concern. Pass an `idempotencyKey` to make retries saf

Approve a pending alt-text suggestion

Approve a PENDING AI-generated alt-text suggestion. Use this when the user reviews a suggestion (typically found via `list_alt_text` with `status=PENDING`) and wants to accept it as-is or with edits. Omit `approvedText` to accept verbatim (transitions to APPROVED). Supply `approvedText` to override the AI text with your own (transitions to EDITED). Optional `longDescription` for complex images (charts, infographics). Already-decided rows return `already_decided` without re-mutation. Once approve

Reject a pending alt-text suggestion

Reject a PENDING AI-generated alt-text suggestion. Use this when the user judges a suggestion unsuitable (typically reviewing via `list_alt_text` with `status=PENDING`). Requires a `reason` code (DECORATIVE, ALREADY_MEANINGFUL, NEEDS_LONG_DESC, BRAND_SPECIFIC, OTHER) — feedback is used to improve the generator. Already-decided rows return `already_decided` without re-mutation. Rejection does NOT delete the row; the suggestion stays in REJECTED status as audit history. Requires `alt-text:approve`

Declare AI-generated content

Declare new AI-generated content under the organization. Use this to record chatbots, generated images/audio/video, or public AI text so the required Article 50 transparency label can be rendered. Provide `embedScriptId`, `contentRef` (a stable DOM/media identifier), `contentType`, and `obligation`. Optional `provenance` attaches C2PA-style metadata. Requires `ai-act:write`.

Confirm or dismiss an AI content candidate

Confirm or dismiss an AI content item. Use `action=confirm` to transition a PENDING or LABELED item to CONFIRMED (this activates/renders the transparency label). Use `action=dismiss` with a `rationale` to mark a detection as a false positive or otherwise not requiring a label. Already-dismissed items cannot be confirmed or dismissed again. Requires `ai-act:write`.

Claim editorial-review exemption for AI content

Claim an Article 50(4) editorial-review exemption for an AI content item, typically publicly deployed AI-generated text. Provide `subjectToEditorialReview` plus either `reviewedByNaturalPerson` or `editorialControl`, the responsible person's name/role, and optional evidence. Requires `ai-act:write`.