ImageSorcery MCP is a self-contained, computer-vision-based MCP server that empowers AI assistants with a rich, locally-running image processing toolkit. It exposes tools for cropping, resizing, rotating images, removing backgrounds, drawing text and shapes, and applying logos or watermarks. It includes object detection using state-of-the-art models and Optical Character Recognition (OCR) to extract text from images. The server supports a broad range of pre-trained models for object detection and OCR, and can operate entirely offline, ensuring user data stays on the device. The system consists of a configurable MCP server with a set of tools, prompts, and resources, and can run in multiple transport modes (STDIO, streamable HTTP, SSE). It provides post-install scripts to download models (e.g., YOLO models, CLIP), creates a config.toml, and sets up a models directory and model descriptions. It can be integrated with MCP clients (Claude, Cursor, Cline) and configured via mcpServers, with autoApprove lists of available tools and a configurable timeout. The README also describes privacy/telemetry, installation, and how to run the server locally.