These commands check a site’s pages against WCAG 2.1 and 2.2 at levels A and AA, keep a list of the issues found, and keep the site’s accessibility statement in step with that list. The examples use the made-up site example.com, and the values in them are invented to show the shape of an answer.

Automated checks find only part of accessibility problems. A check with no findings does not show that a site meets the requirements. What Supervisor checks with lists the tools behind a check and what each can and cannot find.

start_accessibility_check

Starts a check. One browser opens the site’s pages at a desktop and a mobile size and runs the checkers on each page. An agent starts a check when a site has no issues listed yet, when the last check is old, or to verify a fix.

Starting from url, the check follows the site’s links up to max_pages. It only browses: it fills no forms, does not sign in, and skips sign-out and add-to-cart links. With journey_id it runs a journey once and checks the page after each step that passes, which reaches carts, signed-in pages and dialogs that a crawl never sees.

  • Inputs: give url or journey_id. urls lists pages to check besides the crawl. max_pages is how many pages to check at most, 50 if left out and within the site’s own limit. devices picks desktop, mobile or both (both if left out). crawl set to false checks only the pages you listed. note labels the check.
  • Returns: the run_id, the site, the settings, what was charged or how many included checks are left today, and how long the check is likely to take.
  • Cost and effect: a check is free, included in the site’s plan or paid with credits, depending on the site. get_accessibility_statement says under next_check what the next one costs, and the agent should ask you before a check that costs credits. Paid with credits, a check of up to 30 pages costs 5 credits and each further page 0.15, and credits for pages that were not checked are returned. The domain does not need to be verified. One check per site runs at a time.
  • What changes: findings become issues that stay from check to check. An earlier issue that the check no longer finds on a page it covered is marked fixed. After the check, the site’s publishing policy may update the accessibility statement.

Example

“Check the accessibility of example.com.”

{
  "run_id": "77777777-8888-4999-8aaa-bbbbbbbbbbbb",
  "domain": "example.com",
  "url": "https://example.com/",
  "status": "running",
  "settings": { "max_pages": 50, "devices": ["desktop", "mobile"], "crawl": true },
  "next": "One browser runs up to 50 pages (about 25 minutes) on desktop and mobile. get_accessibility_check shows progress and results."
}

get_accessibility_check

Shows how far a check has got and, when it is done, what it found. An agent calls it about every 30 seconds while a check runs, then reads the result to you.

  • Inputs: run_id. Set screenshots to false to leave the images out.
  • Returns: the status (running or done), a one-line summary, the pages checked and any that could not be checked, the issues found by WCAG criterion and how many are new, fixed_by_this_check (earlier issues this check no longer found), still_fixed, what happened to the statement (publish), the ten most serious open issues, and engine, the names and versions of the checkers that ran. When the AI review ran, ai_judges says how many pages it read.
  • Images: up to three screenshots of failing elements.
  • Cost and effect: free. It only reads.

Example

“Is the check done? What did it find?”

{
  "run_id": "77777777-8888-4999-8aaa-bbbbbbbbbbbb",
  "status": "done",
  "pages_checked": 42,
  "summary": "Checked 42 pages on desktop and mobile: 18 issues found (5 new). 3 issues that earlier checks found are gone and now fixed.",
  "found": { "issues": 18, "new": 5, "by_criterion": { "1.4.3": 9, "4.1.2": 6, "2.4.7": 3 } },
  "fixed_by_this_check": 3,
  "publish": { "policy": "auto_add", "added": ["1.4.3"], "removed": [], "outcome": "published" },
  "top_issues": [
    { "issue_id": "88888888-9999-4aaa-8bbb-cccccccccccc", "criterion": "1.4.3", "status": "open", "rule": "color-contrast", "impact": "serious", "url": "https://example.com/", "device": "mobile", "elements": 4 }
  ]
}

list_accessibility_issues

Lists a site’s accessibility issues. This is the working list: an agent reads it to decide what to fix first, and it is where the accessibility statement gets its content.

  • Inputs: url is required. status picks which statuses to list. Left out, it lists the ones that still need work: open, needs review and confirmed. criterion narrows the list to one WCAG success criterion, for example 1.4.3. limit is how many to return, up to 200 (50 if left out).
  • Returns: counts by status and by criterion, and issues. An issue has its WCAG criterion, status, page, device, the rule and the checker that found it, the impact, a help text, the element’s selector and HTML, a screenshot link, and when it was first and last seen.
  • Statuses: open (found by a check), needs_review (a check found it, but the rule needs a person to confirm it), confirmed (a person agrees, or added it), false_positive, fixed (a check no longer finds it, or a person closed it) and accepted (fixing it would be an unreasonable burden).
  • Cost and effect: free. It only reads.

Example

“Which contrast problems are still open on example.com?”

{
  "domain": "example.com",
  "showing": ["open", "needs_review", "confirmed"],
  "counts": { "open": 12, "needs_review": 6, "fixed": 9 },
  "issues": [
    {
      "issue_id": "88888888-9999-4aaa-8bbb-cccccccccccc",
      "criterion": "1.4.3",
      "status": "open",
      "origin": "automatic",
      "url": "https://example.com/",
      "device": "mobile",
      "rule": "color-contrast",
      "impact": "serious",
      "elements": 4,
      "selector": ".price-note"
    }
  ]
}

update_accessibility_issue

Sets an issue’s status or adds a note to it. An agent uses it to settle the issues marked “needs review”: it looks at the element and its surroundings, then marks the issue confirmed, or a false positive with a note that says why. It should ask you when it is unsure.

  • Inputs: issue_id is required. status is one of open, confirmed, false_positive, fixed or accepted. note explains the decision.
  • Returns: the issue after the change, with who changed it.
  • Cost and effect: free. It changes data in Supervisor only. A decision made this way sticks: a later check never reopens an issue marked as a false positive or as accepted.

Example

“That logo link is fine, the alt text names the shop. Mark it as a false positive.”

{
  "issue_id": "99999999-aaaa-4bbb-8ccc-dddddddddddd",
  "criterion": "1.1.1",
  "status": "false_positive",
  "note": "The alt text names the shop, which is where the link goes.",
  "changed_by": "agent (owner@example.com)"
}

add_accessibility_issue

Adds a problem that automated checks cannot find, such as a keyboard trap, unclear instructions or missing captions. An agent uses it after testing those things with you, so that the problem is tracked and shows in the statement.

  • Inputs: url (the page), criterion (a WCAG 2.1 success criterion at level A or AA, for example 2.1.2) and note (what fails and where) are required. device, state (for example “cart open”) and selector say more precisely where.
  • Returns: the new issue. Its status is confirmed and its origin is manual.
  • Cost and effect: free. It changes data in Supervisor only. Checks never close a manual issue. Only a person does, with update_accessibility_issue or in the app.

Example

“The video on the front page has no captions. Record that.”

{
  "issue_id": "aaaaaaaa-1111-4222-8333-444444444444",
  "criterion": "1.2.2",
  "status": "confirmed",
  "origin": "manual",
  "url": "https://example.com/",
  "note": "The product video on the front page has speech and no captions."
}

get_accessibility_statement

Reads where a site’s accessibility statement stands. An agent calls it before a check, to learn what the check will cost, and after fixes, to see what the statement would say now.

  • Inputs: url.
  • Returns: the published version and the WCAG criteria it lists as failing, the draft made from the current open and confirmed issues, what the draft would add or remove, whether anything waits for review, the publishing policy, the public statement_url, the site’s plan, whether it can publish (can_publish), what the next check costs (next_check) and the site’s limits, such as the most pages one check may cover.
  • Cost and effect: free. It only reads.

Example

“What does our accessibility statement say now, and what would change?”

{
  "domain": "example.com",
  "policy": "auto_add",
  "agents_may_publish": false,
  "statement_url": "https://accessibility.supervisor.com/example.com/.../statement",
  "published": { "version": 4, "published_at": "2026-11-01T03:20:00Z", "criteria": ["1.4.3", "2.4.7", "4.1.2"] },
  "draft": { "criteria": ["1.4.3", "4.1.2"] },
  "added": [],
  "removed": ["2.4.7"],
  "waiting_for_review": true,
  "can_publish": true,
  "next_check": { "credits": 0, "note": "Included in the plan: 4 of 5 checks left today." }
}

get_accessibility_report

Reads a site’s monthly accessibility report. The report comes with an accessibility plan and is made once a month, after the scheduled check. An agent uses it to tell you what changed during the month.

  • Inputs: url is required. month picks a calendar month, written like 2026-12. Left out, you get the latest report. language is en or fi. Left out, it is the language of the site’s statement.
  • Returns: text, the report in words; data, the numbers as they were when the report was made; a summary; and months, the months that have a report. The report covers what was checked since the previous report, new, fixed and open issues by seriousness and WCAG criterion, the state of the statement, feedback received, and the most serious open issues with their pages.
  • Cost and effect: free. It only reads. A site without a plan has no report, and the answer says so. For the state right now, use list_accessibility_issues and get_accessibility_statement.

Example

“What does this month’s accessibility report say?”

{
  "domain": "example.com",
  "month": "2026-11",
  "language": "en",
  "summary": "...",
  "text": "...",
  "months": ["2026-11", "2026-10"],
  "note": "Automated checks find only part of accessibility problems: the report does not show that the site meets the requirements."
}

publish_accessibility_version

Publishes the draft as a new version of the public accessibility statement, removals included. An agent may do this only when the site’s owner has allowed agents to publish, in the statement settings on accessibility.supervisor.com, and it should still ask you first every time.

  • Inputs: url.
  • Returns: the statement as it is after publishing, and published_version, the new version’s number.
  • Cost and effect: it changes a public page. Publishing needs can_publish to be true, which may need a plan. If the owner has not allowed agents to publish, or the draft says the same as the published statement, nothing is published and the answer says why.

Example

“The focus problems are fixed and verified. Publish the new statement.”

{
  "domain": "example.com",
  "published": { "version": 5, "published_via": "mcp", "criteria": ["1.4.3", "4.1.2"] },
  "waiting_for_review": false,
  "published_version": 5
}

set_accessibility_policy

Sets when checks update the statement by themselves. An agent changes it only when you ask.

  • Inputs: url and policy, which is one of:
    • auto_add, the default: new problems are published after each check, and removing a criterion waits for a person.
    • review: every change waits for a person.
    • auto: every change is published, removals too.
  • Returns: the site, the policy now in use and whether agents may publish.
  • Cost and effect: free. It changes a setting in Supervisor. Under auto_add a criterion is never taken off the statement unless a person publishes that change.

Example

“I want to approve every change to the statement myself.”

{ "domain": "example.com", "policy": "review", "agents_may_publish": false }

schedule_accessibility_check

Sets up a check that runs by itself: monthly, weekly or daily. Each scheduled check works like start_accessibility_check, updates the issues, and under the publishing policy brings new problems to the statement. An agent sets one up so that the statement follows the site as it changes.

  • Inputs: give url or journey_id. max_pages and devices are as in start_accessibility_check. frequency is monthly (the default, on day 1), weekly or daily. hour, weekday, day_of_month and time_zone set the time, as in schedule_load_test. notify_emails lists up to five addresses that are told when a check cannot start.
  • Returns: the schedule, with its schedule_id, when in words, the next run time and a note that says what its checks cost.
  • Cost and effect: a scheduled check costs the same as a check you or your agent starts, each time it runs: 5 credits for up to 30 pages and 0.15 for each further page, and credits for pages that were not checked are returned. The domain does not need to be verified. Change or remove the schedule with update_schedule and delete_schedule.
  • Not the same as the subscription: Supervisor Accessibility Compliance is a separate product with a fixed monthly price. It includes a scheduled check every month together with other things, such as the accessibility statement page, the feedback form and the monthly report. Pricing shows both ways to pay.

Example

“Check example.com every month and keep the statement up to date.”

{
  "schedule_id": "bbbbbbbb-2222-4333-8444-555555555555",
  "kind": "accessibility",
  "domain": "example.com",
  "when": "monthly on day 1 at 03:00 UTC",
  "enabled": true,
  "next_run_at": "2026-12-01T03:00:00Z",
  "accessibility": { "max_pages": 50, "devices": ["desktop", "mobile"] }
}