ImageSorcery MCP is a local image recognition and editing tool server based on computer vision, mainly aimed at AI assistants, providing image processing capabilities to AI through the MCP protocol.
In terms of use, it enables AI assistants to perform various image processing tasks locally, such as cropping, scaling, and rotating images, removing backgrounds, drawing text and shapes on images, adding logos and watermarks, detecting objects using advanced models, and extracting text from images via OCR. All operations can be completed locally without sending images to a server, helping to protect privacy.
Main capabilities include: precisely cropping, resizing, and rotating images; removing backgrounds; drawing text and shapes on images; adding logos and watermarks; performing object detection using pre-trained models; extracting text via OCR; supporting object detection and text prompt search using multiple pre-trained models (such as YOLO, CLIP). Users can issue natural language instructions to the AI, such as "copy photos with pets from the folder photos to the folder pets", "find the cat in photo.jpg and crop the image to center it", etc.
In terms of usage, Python 3.10 or higher is required, and installation with pipx is recommended. System libraries such as ffmpeg, libsm6, libxext6, and libgl1-mesa-glx are also required, as well as Claude.app, Cline, or another MCP client. The installation command is pipx install imagesorcery-mcp, after which run imagesorcery-mcp --post-install to download the required models and install the clip package. A virtual environment can also be created with python -m venv and then installed via pip. After installation is complete, configuration must be added in the MCP client. The server supports three operating modes: STDIO (default), Streamable HTTP, and SSE.
Applicable scenarios include: having an AI assistant help batch process images, automatically detect and edit image content, extract fields from form images and fill in documents, etc.
In terms of limitations: file paths are unrestricted by default, but the allowed directories can be restricted by setting IMAGESORCERY_AVAILABLE_PATHS. Telemetry is disabled by default and requires the user to explicitly opt in to enable it. When using uv venv or uvx, the clip package may not be installed automatically; it is recommended to use python -m venv. The project uses the MIT license.
