varunr89/claude-marketplace/image-ocr

Image OCR

Batch-processes a directory of JPG/JPEG images and produces a combined Markdown file containing the OCR text from each image.

When to use

Use this skill when the user wants to extract text from one or more JPG/JPEG images -- for example scanned documents, photos of whiteboards, or screenshots.

Backends

• vision (default on macOS/Apple Silicon) -- Uses Apple's Vision framework via PyObjC. Fast, on-device, GPU-accelerated. Requires pyobjc-core and pyobjc-framework-Vision.
• tesseract -- Uses the tesseract CLI binary. Cross-platform. Optional preprocessing (grayscale, autocontrast, upscale) via Pillow.
• mlx_ocr -- Uses MLX-based OCR on Apple Silicon. Runs in a subprocess to isolate potential Metal crashes. Requires Python 3.10+ and mlx-ocr.
• auto -- Tries vision first, falls back to tesseract.

Usage

python3 ${CLAUDE_PLUGIN_ROOT}/scripts/ocr_jpgs_to_markdown.py \
  --input <directory-of-jpgs> \
  --output <output.md> \
  [--backend auto|vision|tesseract|mlx_ocr] \
  [--start N] [--limit N] \
  [--languages en-US] \
  [--fast] \
  [--workers N] \
  [--tesseract-lang eng] [--tesseract-psm 6] [--tesseract-oem 1] \
  [--preprocess] [--preprocess-scale 2.0] \
  [--det-lang eng] [--rec-lang lat] \
  [--mlx-worker-python /path/to/python3.11]

Key arguments

Dependencies

• macOS + PyObjC for Vision backend (pip install pyobjc-core pyobjc-framework-Vision)
• tesseract binary for tesseract backend (brew install tesseract)
• mlx-ocr + Python 3.10+ for MLX backend
• Pillow for preprocessing and MLX