Local execution is the distinguishing constraint: detection, OCR and editing all run on your own machine, so photographs you are not allowed to upload are still workable. Asking in plain language — find the cat and crop around it — chains several tools in one go.
A local computer-vision toolkit for AI assistants, built on OpenCV. It crops, resizes and rotates, removes backgrounds, draws text and shapes, adds logos and watermarks, detects objects with pretrained models and extracts text with OCR — all on the machine it runs on.
- Crop, resize and rotate with precision, plus background removal
- Object detection with a range of pretrained models, chosen per call with a confidence threshold
- Text extracted from images by OCR
- Rectangles, circles, lines and arrows drawn onto an image (draw_rectangles, draw_circles, draw_lines, draw_arrows)
- Colour changes applied to an image (change_color)
- Text, logos and watermarks added
- Several tools chained by one request, so a single instruction can find, crop and annotate
- Everything runs locally — images are never sent to a server
A local install of the imagesorcery-mcp package. Give it full paths to your files — the tools read and write images on the machine. MIT licensed, and it works with Claude, Cursor and Cline.
One command — pipx install imagesorcery-mcp