Process images and videos with OpenCV

An MCP server that gives AI assistants OpenCV computer vision capabilities for processing images and video: object detection, face recognition, edge detection, and video analysis, exposed as a set of chainable tools.

Use this when an AI assistant needs to manipulate or analyze images or video directly - resizing, filtering, detecting edges, contours, or shapes, matching templates or features, detecting faces or objects, or extracting and analyzing video frames. Object and face detection require pre-trained models (YOLO weights, Haar cascade, or DNN face-detection models) placed in the models directory first - it is not a ready-to-detect solution without supplying those model files yourself.

• Basic image handling: save_image_tool, convert_color_space_tool (BGR/RGB/GRAY/HSV and more), get_image_stats_tool.
• Image processing: resize_image_tool, crop_image_tool, apply_filter_tool (blur, Gaussian, median, bilateral), apply_threshold_tool.
• Edge and shape analysis: detect_edges_tool (Canny, Sobel, Laplacian, Scharr), detect_contours_tool, find_shapes_tool (circles, lines).
• Matching and features: match_template_tool, detect_features_tool (SIFT, ORB, BRISK), match_features_tool.
• Detection: detect_faces_tool (Haar cascades or DNN), detect_objects_tool (pre-trained DNN models such as YOLO).
• Video processing and analysis (frame extraction, motion detection), object tracking in video, and camera integration for real-time object detection.
• Tools can be chained together by feeding one tool's output_path into the next tool as its input.

Install with pip install opencv-mcp-server, or via uvx (install uv first with brew install uv), or from source: clone the repo, create and activate a virtual environment with python -m venv .venv and source .venv/bin/activate (or .venv\Scripts\activate on Windows), then run pip install -e .. For Claude Desktop:

{
    "mcpServers": {
        "opencv": {
            "command": "uvx",
            "args": [
                "opencv-mcp-server"
            ]
        }
    }
}

Optional environment variables: MCP_TRANSPORT (transport method), OPENCV_DNN_MODELS_DIR (DNN model directory), and CV_HAAR_CASCADE_DIR (Haar cascade file directory).

Typical prompts it handles include "Resize an image to 800x600 pixels", "Apply Gaussian blur filter to reduce noise in an image", "Detect faces in a group photo", "Find objects in a street scene using the YOLO model", and "Extract every 10th frame from a video".

Developers building AI assistants or agents that need direct image and video processing - from basic resizing and filtering to face and object detection - without writing custom OpenCV integration code from scratch, and who are comfortable supplying their own pre-trained detection models rather than relying on ones bundled with the server.