ImageSorcery MCP
A local MCP server for editing and analyzing images: the agent crops frames, finds objects and reads text from photos
- 334 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The agent reads and writes image files at the given paths, so keep a backup of your originals
- The install downloads models and the clip package from a GitHub repository
Install
- Works in
Manual install
pipx install imagesorcery-mcp && imagesorcery-mcp --post-installThe second step is required: it downloads the models and installs clip for the find tool.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The agent reads and writes image files at the given paths, so keep a backup of your originals
- The install downloads models and the clip package from a GitHub repository
This is third-party code. Review the repository files before installing.
What it does
ImageSorcery MCP gives an agent a set of image operations that run on your own computer. Pictures are not sent to third-party servers. The work is built on OpenCV, Ultralytics YOLO models and EasyOCR. For editing there are crop, resize and rotate. The server also blurs and fills areas, overlays one image on another and changes colors. For annotation the agent adds text, shapes and arrows to a picture. Recognition finds objects with the chosen model or from a text description, and OCR pulls text out of a picture. Separate tools read file metadata and change server settings. The agent chains the operations itself. You can ask it to find a cat in a photo and crop the frame so the cat ends up centered, or to sort pet photos into folders. The ready remove-background prompt walks the agent through detection and masks. The models://list resource returns the installed models. The server runs over stdio by default, and streamable-http and sse modes are switched on separately.
Who it is for. For people who process photos and screenshots through Claude Desktop, Cline and other MCP clients and do not want to send files to the cloud.
Good fit when
- You need to sort a folder of photos and copy the frames with animals into another folder
- You need to crop, shrink or rotate a batch of images from a description, with no manual editor
- You need text from a screenshot or a scan through OCR in the right language
- You need to paint over or blur a region of an image, such as faces or license plates
- You need to cut out an object from a text description and remove the background
Not a fit when
- You want to generate pictures from scratch: ImageSorcery edits and analyzes existing images
- You need heavy retouching and color grading at the level of a graphics editor: the tools cover simple OpenCV operations
- You have no Python 3.10 or cannot install the OpenCV system libraries, for example in a minimal container
- You lack disk space or time to wait for model downloads on the first install
Example request
Find the cat in photo.jpg and crop the frame to half its height and width so the cat is centered. Use imagesorcery.MCP
- Transport
- stdio, http, sse
- Authentication
- not required
Limitations
Python 3.10 or later, pipx and the system libraries `ffmpeg`, `libsm6`, `libxext6` and `libgl1-mesa-glx` are required. Without the post-install step the YOLO and CLIP models are not downloaded, and the find tool does not work without CLIP.
In an environment created with `uv venv`, and when run through uvx, the clip package may fail to install. The author recommends pipx or a plain `python -m venv`.
Give files by full path. If you need a particular model, name it and the confidence threshold in the request. Telemetry is off by default and does not collect images, file paths or IP addresses.
How to disable. Remove the imagesorcery-mcp block from the MCP client settings and, if you like, run `pipx uninstall imagesorcery-mcp`. The downloaded models live in the models directory next to the package, and `config.toml` sits where you ran the install.
FAQ
How do I install ImageSorcery MCP in Claude Desktop?
Run `pipx install imagesorcery-mcp` and `imagesorcery-mcp --post-install`, then add a server with the command imagesorcery-mcp and transportType stdio to the settings file. After the client restarts, the tools show up in its list. On Windows the command is `imagesorcery-mcp.exe`.
What does `imagesorcery-mcp --post-install` do?
It creates `config.toml`, prepares the models directory, downloads yoloe-11l-seg, yoloe-11s-seg and the pf variants, and tries to install the clip package and the CLIP model for the find tool. You can run it again to restore the default models.
How do I add another object detection model?
The `download-yolo-models` command fetches more models: use `--ultralytics` for Ultralytics models and `--huggingface` for Hugging Face ones. After downloading, check the descriptions in the models/model_descriptions.json file.
Does ImageSorcery send images to the internet?
No, images are processed on your computer. Anonymous telemetry turns on only with explicit consent and collects the names of the tools used, the version, the OS type and failures.
How does find differ from detect?
The detect tool looks for the classes the chosen Ultralytics model knows. The find tool takes a text description such as dogs and locates matching objects through CLIP. Both can return PNG masks or polygons.
Discussion
No questions or reviews yet.
Related
MCP serversAn MCP server for Adobe Photoshop: background removal, retouch, social export and batch jobs by chat
Recraft's remote MCP server: an agent draws raster and vector images, edits a region, removes the background and upscales
An MCP server for Windows that drives an installed Photoshop: documents, layers and arbitrary JSX scripts
The official remote MCP server of Kling AI: an agent makes videos and images from text and from a picture right in the work chat
Claude Video Vision
claude-video-vision
An MCP server and Claude Code plugin that let the agent watch video: frames via ffmpeg and audio transcription via Gemini, local Whisper or OpenAI
An MCP server for generating images and video from the agent: several models, a large prompt library and parallel tasks
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| ImageSorcery MCP (this entry) | no | not required | 334 | |
| PaddleOCR agent skills | no | 91,959 | narrow skills for text recognition and document parsing, while ImageSorcery adds editing and object search | |
| MeiGen AI Design MCP | yes | API key | 1,793 | creates images and video with cloud models, while ImageSorcery edits existing files locally |
README in short
Each tool, resource and prompt has its own description in the project folders with details and example requests. Cline users get a ready install brief to paste as a request.