scanner

A full-screen, menu-driven terminal scanner for the Brother MFC-L5850DW on macOS. You run it in Terminal and drive a numbered menu (BIOS / green-screen style); there are no flags to memorize. It wraps scanline, which reaches the scanner through the same macOS service Image Capture uses, so talking to the scanner over Wi-Fi is handled by the OS rather than by us.

Built for a bulk job: scanning a large archive of mixed single- and multi-page documents into tidy PDFs.

What the menu does

• Scan documents (batch loop): each feeder load becomes one multipage PDF. Load a document, tap ENTER, it saves label_001.pdf, then it is ready for the next load. Tap, load, tap. This is the flow for multi-page records.
• Split pages: feed a pile and get one PDF per sheet (label_pNNN.pdf). This is the flow for a stack of single-page items.
• Flatbed: a single sheet on the glass.
• Settings: two-sided (duplex), resolution, color vs black and white, blank-page dropping and its sensitivity, and the save folder.

Blank-page dropping

Single-pass duplex scans both sides of every sheet, so single-sided pages come out with a dead blank back. With blank-dropping on (the default), those empty backs are removed automatically. It is deliberately conservative: it only drops pages that are essentially free of ink, so a sparse but real page (a short invoice, a signature page) is always kept. Pages are never re-encoded and a page with real content is never deleted. The tool reports which page numbers it dropped after each scan.

The sensitivity is a percent-of-ink threshold (default 0.15). Truly empty backs measure near 0.00 percent; real pages measure roughly 0.3 percent and up. Raise the threshold to drop more aggressively, lower it to keep more.

Batch naming

In Scan documents, each feeder load is named YYYYMMDD-n automatically (for example 20260716-1, 20260716-2); there is no name to type. The next name is shown before you scan and the saved name after. Numbering continues past any file already saved that day, so a second batch on the same day never overwrites the first.

Status

Verified on the real scanner, end to end:

• scanline installed and reaching the L5850DW (scanline -list shows it).
• All scanline flags used are confirmed against scanline -help.
• Flatbed and duplex feeder scans produce correct, correctly-ordered PDFs.
• A full Scan documents batch run with blank-drop on was verified page by page: it dropped the empty duplex backs, kept every page with content (including a sparse invoice and both sides of a genuinely double-sided document), and produced a correct multipage PDF. Ready for the bulk job.

Requirements

• macOS with Python 3 (already present on current macOS). Standard library only.
• scanline: download the signed, notarized installer from https://github.com/klep/scanline and install it (it places a scanline binary in /usr/local/bin).
• poppler: brew install poppler. Provides pdfseparate, pdfunite, and pdftoppm, used to split, recombine, and blank-probe pages without re-encoding them.
• macOS sips (built in), used only for the tiny blank-probe images.

Note on the SANE / scanimage backend: it is not a working fallback on this machine. The Homebrew build of sane-backends has no eSCL backend, so scanimage cannot reach a Wi-Fi Brother, and sane-airscan is not in Homebrew. scanline is the backend here.

The scanner must appear in Apple's Image Capture app first; that is the quick proof macOS can drive it. AirPrint working does not prove scanning works; scanning is a separate service (AirScan / eSCL).

Running it

python3 scan.py            # opens the full-screen menu
python3 scan.py --dry-run  # prints the commands it would run, scans nothing

--dry-run walks the menus and prints the exact commands each option would run, with no scanner and no backend needed.

How it works

• scan.py is the whole tool: the screen UI, settings, and the scan pipeline.
• Settings persist to ~/.config/scanner-tui/config.json, so you answer the prompts once, not every time.
• Scans are saved under ~/Documents/Scans/<date>/.
• Each scan runs scanline to a temporary native PDF (its page ordering is correct and proven). If nothing needs filtering, that PDF is moved into place as-is. Otherwise it is split per page with pdfseparate, each page is probed for blankness, and the keepers are recombined with pdfunite (multipage) or saved one per page. No page images are re-encoded.

Handling a mixed stack

Nothing on a plain sheet marks where one document ends and the next begins, so a mixed pile cannot be auto-split into per-document PDFs. Two workable approaches:

1. Sort into a "single-page" pile and a "multi-page" pile. Run the singles through Split pages; run each multi-page document through Scan documents, one document per feeder load.
2. Feed each document as its own load in the Scan documents batch loop. The act of loading one document at a time is what marks the boundary.