PDF Accessibility Checker & Fixer: PDF/UA, WCAG, Redaction

Three checks every PDF should pass before it is published or sent out, in one run:

1. Accessibility: 32 rules in 7 categories (document, page content, forms, alternate text, tables, lists, headings). Each rule passes, fails (with the page and the problem) or needs a person's review. A readable HTML report is saved per PDF. Useful for the European Accessibility Act, WCAG and PDF/UA work.
2. Fake redactions: black boxes painted over text that can still be selected and copied, and redaction marks that were never applied. The row shows the exposed text so you can act on it.
3. Hidden data: metadata, attachments, comments, form fields, hidden text and layers, bookmarks, links, scripts and private application data, counted per category.

Optional fixes save a corrected copy (the original is never changed): title shown in the viewer's title bar, document language, tab order, automatic tagging of untagged text-only PDFs, and removal of hidden data.

Input

{
  "pdfs": ["https://example.com/annual-report.pdf", "https://example.com/court-filing.pdf"],
  "checkAccessibility": true,
  "checkRedactions": true,
  "checkHiddenData": true,
  "fixAccessibility": false,
  "removeHiddenData": false
}

Up to 100 PDFs per run, or upload one in pdf. Leave both empty to try a built-in sample PDF for free (its row has "status": "sample"). For encrypted PDFs, give password (stored encrypted). title and language are used by the accessibility fix.

Output

One dataset row per PDF:

{
  "fileName": "court-filing.pdf",
  "status": "checked",
  "pages": 12,
  "tagged": true,
  "accessibility": {
    "passed": 24, "failed": 2, "manual": 5, "skipped": 1,
    "failedRules": [{"rule": "title", "name": "Title", "category": "document",
                     "findings": [{"message": "The document has no title", "page": null}]}],
    "manualReview": ["Logical Reading Order", "..."]
  },
  "redactions": {"count": 1, "findings": [{"page": 3, "kind": "unapplied-mark", "coveredText": "John Smith"}]},
  "hiddenData": {"total": 10, "categories": {"metadata": 5, "bookmarks": 4, "links-actions-scripts": 1}},
  "reportUrl": "https://api.apify.com/v2/key-value-stores/.../records/002-court-filing-accessibility.html",
  "detailsUrl": "https://api.apify.com/v2/key-value-stores/.../records/002-court-filing-details.json",
  "fixedPdfUrl": null,
  "error": null
}

With fixes on, fixes lists what was applied, any warnings, and the accessibility result of the fixed copy, and fixedPdfUrl links to it.

Use with AI agents (MCP)

https://mcp.apify.com?tools=shatrix/pdf-compliance-checker

Use from code

curl -X POST "https://api.apify.com/v2/acts/shatrix~pdf-compliance-checker/run-sync-get-dataset-items" \
  -H "Authorization: Bearer $APIFY_TOKEN" -H "Content-Type: application/json" \
  -d '{"pdfs": ["https://example.com/report.pdf"]}'

Pricing

Pay per event: one charge per checked PDF, and one per fixed copy (see the Pricing tab). Files that can't be opened (not a PDF, wrong password) are not charged.

Important

• Automatic checks find many, not all, accessibility problems; rules marked for manual review need a person. A passing result is not a certification of PDF/UA or WCAG conformance, and this is not a PDF/UA validator: on veraPDF's PDF/UA test files it flagged untagged content, but not a missing PDF/UA identifier or fonts that aren't embedded.
• Automatic tagging tags text as paragraphs; review headings and reading order afterwards. It works on text-only PDFs and is skipped (with a warning) for others.
• A PDF whose security settings forbid changes is still checked; fixing it needs its permissions (owner) password in password, otherwise the row explains why there is no fixed copy and only the check is charged.
• Tested with real PDFs: an IRS tax form with 199 fields, a 142-page IRS publication, scanned pages, and AES-128 and AES-256 encrypted files. The redaction audit flagged nothing on these real files.
• Files are processed inside the Actor run and are not sent to any other service.

Credits

Built on PdfCraft from the ArtCraft project (MIT OR Apache-2.0). Text recognition models: ocrs-models by Robert Knight, trained on HierText, CC-BY-SA-4.0. License texts ship in the Actor's image under /opt/engines/licenses/. This Actor is not affiliated with or endorsed by the ArtCraft project.