A simple CLI tool for adding structured bookmarks to PDFs.

bmrk analyses a PDF's text and font metadata to detect its heading structure, then writes a bookmarked copy for easier navigation in any PDF viewer.

• Installation
  • From source
  • With OCR support
    • OCR in a dev environment
• Usage
  • Basic
  • Options
  • Inspect before writing
  • Manual heading adjustments
  • Tune for a noisy PDF
  • Handle a cover page
• How it works
• Code structure
• Limitations
• Development
• Contributing
• License

pip install bmrk

For an isolated install that keeps bmrk available globally without polluting your Python environment:

# pipx
pipx install bmrk

# uv
uv tool install bmrk

To run bmrk once without installing it:

# pipx
pipx run bmrk paper.pdf paper_bookmarked.pdf

# uvx (uv's ephemeral tool runner)
uvx bmrk paper.pdf paper_bookmarked.pdf

pip install git+https://github.com/AnvarAtayev/bmrk.git

For scanned PDFs that lack a text layer, install the optional OCR extra:

pip install "bmrk[ocr]"
# or
pipx install "bmrk[ocr]"
# or
uv tool install "bmrk[ocr]"

This pulls in ocrmypdf, which itself requires Tesseract and Ghostscript to be installed on your system:

# macOS
brew install tesseract ghostscript

# Debian/Ubuntu
sudo apt install tesseract-ocr ghostscript

# Windows -- download installers from:
#   https://github.com/UB-Mannheim/tesseract/wiki
#   https://www.ghostscript.com/releases/gsdnld.html

Then pass --ocr to bmrk:

bmrk scanned.pdf scanned_bookmarked.pdf --ocr

# 1. Clone the repo and sync all extras
git clone https://github.com/AnvarAtayev/bmrk.git
cd bmrk
uv sync --extra dev --extra ocr

# 2. Install system deps (macOS example)
brew install tesseract ghostscript

# 3. Run
uv run bmrk scanned.pdf out.pdf --ocr

bmrk [OPTIONS] <INPUT>.pdf [<OUTPUT>.pdf]

bmrk paper.pdf paper_bookmarked.pdf

bmrk paper.pdf --dry-run --verbose

If the auto-detected bookmarks are not quite right, you can export the heading structure, edit it by hand, and import the corrected version back in.

Step 1 -- Export the detected headings

bmrk paper.pdf --export-headings headings.tsv

When OUTPUT is omitted, bmrk runs detection and exports the heading list without writing a PDF.

Step 2 -- Edit the TSV file

Open headings.tsv in any text editor or spreadsheet app. The format is tab-separated with three columns:

# bmrk heading export
# level	page	title
1	1	Introduction
2	3	Background
2	7	Methods
1	12	Results
3	14	Statistical Analysis

• level -- heading depth (1 = top-level chapter, 2 = section, 3 = subsection, ...).
• page -- 1-based page number where the heading appears.
• title -- the bookmark text shown in the PDF viewer.
• Lines starting with # are comments and are ignored on import.

Common edits:

• Remove a heading -- delete the line entirely.
• Add a missing heading -- insert a new line with the correct level, page, and title.
• Fix a title -- change the text in the third column.
• Change nesting -- adjust the level number (e.g. change 2 to 1 to promote a section to a chapter).
• Reorder headings -- rearrange lines; bookmarks are inserted in the order they appear in the file.

Step 3 -- Import and produce the bookmarked PDF

bmrk paper.pdf paper_bookmarked.pdf --import-headings headings.tsv

This skips detection entirely and uses your edited headings to write the bookmarked PDF.

# More conservative -- only large headings
bmrk paper.pdf out.pdf --threshold 1.15

# More aggressive -- catches bold same-size section titles
bmrk paper.pdf out.pdf --threshold 1.01

# Skip page 1 (the cover) when detecting headings
bmrk report.pdf report_bookmarked.pdf --cover-pages 1

bmrk reads every text span in the PDF along with its font size and style, then uses three signals to find headings:

1. Font size -- text larger than the body font is a heading. The biggest text becomes H1, the next size H2, and so on.
2. Numbered prefixes -- lines like 1 Introduction or 2.3 Methods are headings, with depth inferred from the numbering.
3. Bold/italic at body size -- some documents style section headings in bold or italic without changing the font size. These are picked up as the lowest heading level.

After detection, bmrk cleans up the results (removes running page headers, deduplicates, merges chapter labels like Chapter 1 with the title that follows) and writes the final bookmark outline into the output PDF.

flowchart LR
    A[PDF] --> B[Extract spans]
    B --> C[Pre-process]
    C --> D[Detect headings]
    D --> E[Clean up]
    E --> F[Write bookmarks]

    C -.- C1["Skip cover/TOC pages
    Exclude headers/footers
    Estimate body font size"]
    D -.- D1["1. Font size > body size
    2. Numbered prefixes
    3. Bold/italic at body size"]
    E -.- E1["Remove running headers
    Deduplicate adjacent titles
    Merge chapter labels
    Filter by max depth"]

src/bmrk/
├── cli.py        # Typer CLI entry point
├── detector.py   # Heading detection logic and HeadingEntry dataclass
├── bookmarker.py # PDF bookmark writing

• Scanned/image PDFs -- bmrk cannot detect headings in PDFs without selectable text. Run OCR first with bmrk --ocr (requires bmrk[ocr]).
• Existing bookmarks -- bmrk replaces any existing outline; it does not merge with pre-existing bookmarks.

uv sync --extra dev

# Lint
uv run ruff check src/

# Test
uv run pytest

Contributions are welcome. Bug reports, feature requests, and pull requests can all be submitted via GitHub Issues or as a pull request against main.

Before opening a pull request, run the lint and test suite to confirm nothing is broken:

uv sync --extra dev
uv run ruff check src/
uv run pytest

MIT