pdf_annotation_extractor
Extracts annotations (highlights, comments, notes) from a PDF, outputs Markdown or JSON. Intended for sharing feedback in plain text instead of the annotated PDF.
Main problem Text markup in PDFs records where a highlight is, not the text underneath. It needs to be retrieved.
Output example
## Comments
- Page #4:
> Quote text highlighted in the PDF.
Your comment in the annotation.
Installation
cargo install pdf_annotation_extractor
Usage
pdf_annotation_extractor paper.pdf # Markdown
pdf_annotation_extractor paper.pdf -f json # JSON
Output options
--show colour,kind,date adds identifiers beside the page number.
--number global or --number per-page numbers items, so a remark can be cited as "page 3, note 2".
Anything finer (bullet, indent, headings, escaping) is set through markdown::Style, which the library exposes. The CLI uses the defaults.
Library
cargo add pdf_annotation_extractor --no-default-features
default-features = false drops clap, which only the binary uses.
Similar tools
pdfannots tackles the same problem.
pdfannots2json tackles the same problem, focused on JSON output.
pdf-annotations-converter focuses on GoodReader's text export.
pdf-annot reads, creates, and edits PDF annotations. Part of PDFluent
cli-pdf-extract focuses on LLM-friendly PDF inspection.
Tests
We test annotations inserted by hand using various PDF readers. Happy to include more if your case is not covered.
Known issues and limitations
- Right-to-left scripts come out reversed.
Text is assembled by sorting glyphs left to right. Moreover, there is no continuation of reading order by looking just at an annotation. Therefore, this is unsupported.
- Markup without
/QuadPointsyields the comment but no covered text.
The PDF specification requires quads on text markup and no producer considered omits them.
- Reply threads (
/IRT) are not followed, so a reply appears as a separate item./Caretinsertion marks are skipped.
There is no "reply" or "caret" options in many PDF readers, so this is not in scope for now.
- Line-break hyphens are joined, which suits LaTeX multi-column text ("soft-
\nware" to "software") but not always ("memory-\nmapped" to "memorymapped").
These cases cannot be told apart. --keep-hyphens disables the default behaviour.