Inspect PDF tags and language
Read roles, page/MCID links, parent references, language and structured order. Report unknown coverage and human checks; no WCAG/PDF-UA certification.
- Runs in your browser
- No upload
- Check before download
Files stay in this browser. Read roles, page/MCID links, parent references, language and structured order. Report unknown coverage and human checks; no WCAG/PDF-UA certification.
Supported profile
Works with traditional PDF 1.0–1.7 files up to 10 MiB, 20 pages and 10,000 objects. Expanded streams are limited to 4 MiB each and 20 MiB in total, and images to 4 MP each and 16 MP in total. Raster comparison is limited to 4 MP per page and 16 MP in total, the worker stops after 90 seconds, and the output is at most 30 MiB. Encryption, signatures, XFA or scripts, object streams, incremental revisions, indirect stream lengths, inline images and Type3 fonts are not supported. Only bundled local fonts and CMaps are used, and your originals stay unchanged. Image predictors are bounded and matched to the image size, and colors use at most four channels (indexed: up to 256 entries). Only sampled and exponential functions are supported, up to 1,048,576 sampled values; other complex color or function profiles are rejected. Repeated form programs expand to at most 500,000 tokens with recursion up to 32, and booklet copy planning also fits the 20 MiB expansion budget. Forms and bookmarks: up to 200; text edits: up to 1024 characters; bookmark depth: up to 6; annotations: up to 1000. Title and author edits: up to 1024 characters; XMP: up to 256 KiB of UTF-8; the repair structure report: up to 2 MiB. Static non-interlaced PNG: up to 1 MiB and 1024×1024.
How to use
- Choose PDF. Files stay in this browser.
- Read roles, page/MCID links, parent references, language and structured order. Report unknown coverage and human checks; no WCAG/PDF-UA certification.
- Preview. Keep original. Works with traditional PDF 1.0–1.7 files up to 10 MiB, 20 pages and 10,000 objects. Expanded streams are limited to 4 MiB each and 20 MiB in total, and images to 4 MP each and 16 MP in total. Raster comparison is limited to 4 MP per page and 16 MP in total, the worker stops after 90 seconds, and the output is at most 30 MiB. Encryption, signatures, XFA or scripts, object streams, incremental revisions, indirect stream lengths, inline images and Type3 fonts are not supported. Only bundled local fonts and CMaps are used, and your originals stay unchanged. Image predictors are bounded and matched to the image size, and colors use at most four channels (indexed: up to 256 entries). Only sampled and exponential functions are supported, up to 1,048,576 sampled values; other complex color or function profiles are rejected. Repeated form programs expand to at most 500,000 tokens with recursion up to 32, and booklet copy planning also fits the 20 MiB expansion budget. Forms and bookmarks: up to 200; text edits: up to 1024 characters; bookmark depth: up to 6; annotations: up to 1000. Title and author edits: up to 1024 characters; XMP: up to 256 KiB of UTF-8; the repair structure report: up to 2 MiB. Static non-interlaced PNG: up to 1 MiB and 1024×1024.
How to use
- Choose PDF
Open a supported local PDF and inspect the tag structure. Look in the report for the document language, the role hierarchy, the page and MCID links, and the parent references.
- Supported profile
A tagged paragraph P linked to page 1 and MCID 0 should appear as that role and link. Compare its structured order with the actual paragraph on the page.
- Process PDF
Read the missing, unresolved and unknown coverage separately. Check the reading order and meaningful alternative text by hand. Save the structural report, but don't treat it as a repaired PDF.
Supported profile
What does this result mean?
Tags describe the logical structure, which is not necessarily the right meaning or reading order. A clean tree doesn't certify WCAG or PDF/UA accessibility.
How do I resolve rejected or missing input?
If tags or the language are missing, the inspector reports that and can't invent them. Encryption, signatures or object streams that aren't supported need another suitable source or workflow.
What are the limits of this tool?
Local, limited inspection. It does not repair tags or certify accessibility. Traditional PDF 1.0–1.7, ≤10 MiB/20 pages/10000 objects. Originals stay unchanged.