mac-ocr

mac-ocr

mac-ocr recognizes text in images and PDFs and writes searchable PDFs, on macOS via Apple's Vision framework. Two operations:

• OCR (the default action) — recognize text in images and PDFs.
• searchable-pdf — write a PDF with an invisible, selectable text layer, from a PDF or an image (image → one-page PDF sized from embedded DPI, falling back to 72 DPI).

OCR (default action — no subcommand)

OCR is the default, so a file argument alone runs recognition. mac-ocr ocr <file> also works but isn't required.

mac-ocr photo.png                       # text → stdout
mac-ocr scan.pdf                         # multi-page PDF, streamed
mac-ocr a.png b.png c.png                # multiple images
cat screenshot.png | mac-ocr             # stdin (auto-detected)
mac-ocr https://example.com/img.png      # URL (simple GET only)
mac-ocr shots/*.png -o '[dir]/[name].txt'  # a .txt next to each image
mac-ocr --format jsonl scans/*.pdf       # streaming JSONL for big jobs

Flags

Output formats

JSON shape

Each result (one per image, or one per PDF page) is:

{
  "source": { "type": "file", "path": "scan.pdf" },
  "page": 1,
  "pageCount": 3,
  "width": 1224,
  "height": 1584,
  "text": "Full recognized text\nwith newlines",
  "observations": [
    {
      "text": "Full recognized text",
      "confidence": 1.0,
      "boundingBox": { "x": 0.05, "y": 0.42, "width": 0.37, "height": 0.06 },
      "requestRevision": 3
    }
  ]
}

Bounding boxes are normalized 0–1 with a top-left origin. To get pixels: obs.boundingBox.x * result.width, obs.boundingBox.y * result.height.

Exit codes

searchable-pdf

Writes a PDF that looks identical to the source but with selectable, searchable text. By default writes one [name].ocr.pdf per input; pass --merge to combine inputs into one PDF.

mac-ocr searchable-pdf scan.pdf                      # writes scan.ocr.pdf
mac-ocr searchable-pdf photo.jpg                      # image → one-page photo.ocr.pdf
mac-ocr searchable-pdf *.pdf                          # writes <name>.ocr.pdf for each
mac-ocr searchable-pdf scan.pdf -o out/               # out/scan.ocr.pdf
mac-ocr searchable-pdf scan.pdf -o '[name]-ocr.pdf'   # scan-ocr.pdf
mac-ocr searchable-pdf scan.pdf -o -                  # PDF to stdout (refused on a TTY)
mac-ocr searchable-pdf --merge -o lease.pdf page1.jpg page2.jpg

• -o <dest>: path, [name] template, directory, or - for stdout. A fixed path or - takes a single input; multiple inputs need a directory or [name] template.
• --merge combines file/URL inputs into one searchable PDF in exact argument order. It requires -o <file.pdf> or -o -; directory, template, and stdin inputs are rejected. Merged PDFs are rewritten, so annotations/outlines/metadata are not preserved.
• PDF inputs keep their original pages verbatim in non-merge mode (vector content is not re-rasterized); only an invisible text layer is added, and pages that already have selectable text are left untouched. The page is rasterized internally to feed OCR.
• Fully born-digital PDFs pass through byte-for-byte in non-merge mode (annotations/links/forms/outlines preserved). When any page needs OCR or --merge is used, the rewrite preserves page content but not annotations, outlines, or metadata.
• --ocr-all-pages overrides that skip and OCRs every page — needed for hybrid pages (a scan plus a small digital stamp/page number, which counts as "has text"); existing digital text may then appear twice in copy/search.
• Image inputs become one page, sized from embedded DPI metadata when available. Images without usable DPI metadata fall back to 72 DPI (1px = 1pt).
• --image-quality <0–1> controls the visible image layer for image inputs only. OCR still uses the original full-resolution image; PDF inputs are not recompressed.
• --image-page-dpi <36–2400> overrides image input page sizing only. OCR still uses the original full-resolution image; PDF inputs are unaffected. --pdf-dpi remains PDF-page rasterization for OCR.
• --image-downsample-dpi <36–2400> caps the visible image layer resolution for image inputs only. OCR and page size are unaffected; PDF inputs are not downsampled.
• Advanced diagnostics: MAC_OCR_DEBUG=1 draws searchable-PDF OCR boxes into file outputs and writes a JSONL sidecar next to each output PDF (file.pdf → file.jsonl). Use it when diagnosing missing/duplicated text: red accepted line boxes, blue word boxes, orange rejected observations; sidecar records recognition.passes.partitioned, origin, and rejection.reason/supersededBy. Rejected with -o -; schema is diagnostic, not a public compatibility contract.
• --ocr-strategy auto|standard|partitioned controls searchable-PDF OCR strategy. auto is default and may run recursive partitioned OCR for large pages with small detected text; standard forces full-page OCR only; partitioned forces partitioned OCR for eligible pages. Partitioning splits each region along its longer axis with overlap, then keeps splitting only while the region is above the calibrated Vision size floor and text remains small or absent. Very large partitioned runs show an interactive warning after the full-page pass. Auto skips partitioning when --roi is set; forced partitioned mode rejects --roi.
• Accepts the same recognition options as OCR (--fast, -l, -c, --pdf-dpi, --roi, --password, custom words, etc.).
• Status is interactive-only on stderr: a live [page/total] counter + name → path line on a terminal; piped runs are silent on success (errors only) — no quiet flag needed. stdout stays clean for -o -. The ocr command shows the same counter when results aren't streaming to the terminal.

The text layer is word-level: one invisible run per recognized word, positioned from Vision's per-word geometry, so selection rectangles track the printed words. Text is searchable and copyable; width within a word remains approximate.

languages

List the recognition languages supported on this macOS version (one BCP-47 code per line). They apply to both OCR and searchable-pdf.

mac-ocr languages           # accurate recognizer
mac-ocr languages --fast    # fast recognizer's set

Node.js API

The package also exposes a typed, promise-based API (import { ocr, createSearchablePdf, supportedLanguages } from 'mac-ocr') that spawns the binary. Inputs are bytes (Buffer/Uint8Array/ArrayBuffer) — read files or fetch URLs in your own code.

const { text, observations } = await ocr(bytes)          // single image or single-page PDF
for await (const page of ocr.pages(pdfBytes)) { /* … */ } // stream multi-page PDF
const pages = await Array.fromAsync(ocr.pages(pdfBytes))  // …or collect → OcrResult[]
const pdf = await createSearchablePdf(bytes)                    // → Uint8Array (PDF bytes)
const langs = await supportedLanguages()                  // → string[] (ocr + createSearchablePdf)

• ocr() throws if given a multi-page PDF — use ocr.pages().
• Options mirror the CLI: fast, languages, confidence, customWords, languageCorrection (default true), minTextHeight, maxCandidates (ocr only), regionOfInterest ({x,y,width,height} | [x,y,width,height] | "x,y,w,h"), pdfDpi, ocrStrategy, imageQuality, imagePageDpi, and imageDownsampleDpi (searchable PDF only), password, signal (AbortSignal).
• Failures throw MacOcrError with .kind ('usage', 'runtime', 'unavailable', …) and .stderr.

Patterns

Heavy/batch jobs → stream JSONL to disk:

mac-ocr --format jsonl large-docs/*.pdf > results.jsonl   # memory-bounded, one line per page

HTTP with auth/POST/cookies — mac-ocr only does simple GET; fetch upstream and pipe:

curl -H "Authorization: Bearer $TOKEN" https://api.example.com/img | mac-ocr -