LastMilePDF is a free desktop app for cleaning up the tag structure behind PDF/UA and WCAG compliance — built for the accessibility work that auto-taggers leave half-finished, and designed to make the manual pass fast and easy, especially for scanned PDFs.
Configurable shortcuts make tag editing a breeze. Press L to format a list complete with list items, labels, and bodies. Press P to convert a mistaken list into a series of paragraphs. Filter to quickly find Figures or Tables. Everything is at your fingertips—no menus.
Two themes, following your system setting by default. Both meet WCAG 2.1 AA for contrast; the light one goes further, to 7:1 for text.
Universities and nonprofits are under real, growing pressure to make PDFs accessible — course materials, grant reports, public filings, scanned archives. Most of that work is manual tag cleanup, and most institutions doing it are stretched thin on both staff and software budget.
Title II ADA rules, Section 508, and WCAG 2.1 AA obligations apply to campus course content, public-facing documents, and grant deliverables alike. The gap is almost always in the tag tree, not the visible page.
LastMilePDF is free and open source under the MIT license. Use it on as many machines as you need — a disability services office, a library digitization lab, a one-person communications team — without a procurement cycle.
It doesn't OCR or auto-tag — it makes the manual review and cleanup pass fast. Keep using whatever auto-tagger or Acrobat license you already have; LastMilePDF covers the editing experience Acrobat's UI makes painful.
Keyboard-first tag editing, built specifically around the repetitive cleanup work that scanned and auto-tagged PDFs generate.
Reorder, retype, merge, and delete tags with keyboard shortcuts built for doing it hundreds of times in a row — not for occasional use.
Fix table headers and scope automatically instead of hand-tagging every <TH>/<TD> relationship.
Drag a rectangle over the page and press one key to tag what's underneath. Text is cut at the rectangle's edges, so a heading swallowed into the paragraph below comes out as a real heading, and a box around bullets or a hanging-indent reference list becomes a properly structured list.
Drag a rectangle over a scanned table and the rows and columns are guessed for you. Adjust the dividers, or let AI lay out the grid, then build the whole table in one step.
Cut a run of page content in two — a list label from its item, a run-in heading from its paragraph — so each half gets its own tag.
Bulk-correct a mis-tagged role across the whole document in one pass.
Sequence repeated cleanup actions into a saved script, then run the same fix across every document in a batch of scans.
Strip extraneous span/div tags that auto-taggers leave behind, without touching the content that matters.
Fix OCR errors in Actual Text with any AI provider you choose, including one your university hosts. Every change is highlighted for review, with one-click revert.
Tag figures on scanned pages that the auto-tagger skipped, and filter straight to figures when it's time to add alt text.
Narrow the tree to Figures, Tables or Lists, to the tags still missing alt text, to empty ones, or to just the ones an AI fix changed.
Walk the tree automatically at the pace you set — no more holding down the arrow key through a 40-page document.
Compare OCR text against the source image side by side, and highlight any difference between Actual Text and OCR so you can review your own edits.
Artifact full-page decorative figures in one click — the fix for auto-taggers that generate a spurious figure tag on every scanned page.
Run the common PDF/UA and WCAG checks locally, jump straight to each failure, and repair the mechanical ones in a click.
List everything the file marks as an artifact, and tag it back into the tree when it was skipped by mistake.
Reopen a document and it comes back to the tag, the expanded branches and the page you left it on.
Turn on the Desktop Agents connection and a Claude session on the same computer, in the Claude desktop app or Claude Code, can work on the PDF you have open. Ask it to fix one table, every table like it, or the whole document.
The page you're on, the tags you've selected, the whole tag tree, the Verify report and pictures of any page. It turns the page and selects tags to show you what it means.
One request runs a full remediation pass from instructions you can rewrite: heading outline, lists and tables split across pages, captions, alt text, bookmarks, title and author, then a report of what still needs a person.
You watch the edits land, Esc takes control back, and a whole session is one undo. Claude can't open or save files, and only programs on your own computer can connect.
Three of the places the manual pass usually gets slowest: proofreading OCR text against the page, sorting out table headers, and working out what still needs fixing.
It's meant to complement the tools you likely already have — an auto-tagger and/or Adobe Acrobat — by making the manual review and correction pass fast.
Free, MIT-licensed, no account required.
LastMilePDF