ImageSorcery MCP

pip install imagesorcery-mcp

Cropping, resizing, rotating, drawing shapes and text, watermarking, OCR and object detection all run on the machine through OpenCV and locally downloaded detection models, which is the point when the images are private. The design detail that makes it composable is area inversion: blur and fill accept a region and a flag to apply the effect everywhere except that region, so once detection returns the subject, blurring the background is a single further call rather than a mask computed by hand. Detection models are selectable per call with a confidence threshold, so a specialised model can be pointed at a specialised job instead of forcing everything through one general detector.