ImageSorcery MCP: seventeen local image tools wired into an MCP client
An MCP server providing tools for image processing operations
At a glance
- What is it?
- A Python MCP server that hands an AI assistant seventeen image tools, one model listing resource and one background removal prompt, all running through OpenCV, Ultralytics and EasyOCR on your own machine. The install needs OpenCV system libraries, a separate model download step, and a config file you may have to write by hand.
- Who is it for?
- ImageSorcery MCP fits a workflow where an assistant edits files already on your disk and you do not want images uploaded anywhere. Before adopting it, confirm the four OpenCV system libraries exist in your container image, read what amplitude-analytics and posthog report before you repeat the local only claim to anyone, and pin the version yourself: pyproject.toml says 0.12.0 and there are no GitHub releases to roll back to.
- Can I use it commercially?
- Yes. MIT is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Yes. The repository last received commits 180 days ago.
- What is it written in?
- Mainly Python, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on November 15, 2026, and from our analysis. They are not legal advice.
Star history
- sunriseapps/imagesorcery-mcp339 ★ · +9
| Repository | 2026-09-17 | 2026-11-15 |
|---|---|---|
| sunriseapps/imagesorcery-mcp | 330 | 339 |
9 Stars in the last 59 days, measured from the daily GitHub snapshot. Compare with other projects
Editorial analysis
Seventeen tools split across OpenCV, Ultralytics and EasyOCR
The tool table is the substance of this project, and each row names the library behind it. `crop` uses OpenCV's NumPy slicing, `rotate` calls the imutils.rotate_bound function, and `get_metainfo` leans on Pillow, which pyproject.toml describes as being for retrieving image metadata. `detect` runs models from Ultralytics and `ocr` runs EasyOCR, which is where the heavy weight downloads land. The rest are geometry and drawing: `resize`, `change_color`, `overlay`, `draw_arrows`, `draw_circles`, `draw_lines`, `draw_rectangles`, `draw_texts`, `blur`, `fill`, plus `find`, which takes a text description instead of a model name, and `config`, which is the only tool that reads and writes settings rather than pixels. Two flags in pyproject.toml frame the maturity: the version is 0.12.0 and the classifier reads Development Status :: 3 - Alpha. requires-python is >=3.10.
Blur and fill take the same coordinates and can invert them
The interesting design choice is that `blur` and `fill` are the same selection primitive with a different effect, and both can invert what you hand them. The README's own example blurs the area from (150, 100) to (250, 200) with a blur strength of 21, and notes the same tool can invert the provided areas, which is how you blur a background instead of a face. `fill` accepts rectangles, polygons or mask based areas, takes a color with opacity, and can make the selected region transparent, which is the other half of background removal. That pairing is why `remove-background` exists as a prompt rather than a tool. What the documentation never states is the coordinate convention: whether (150, 100) is measured from the top left, and whether the first pair is x and y or row and column. Check it on a scratch image before you trust a coordinate that came out of a conversation.
models://list is the only resource, and it is how you name a model
One resource is exposed, `models://list`, which lists all available models in the models directory. That matters because the detection tools are useless without a model name, and the example prompt in the documentation asks for form fields on a form.jpg with the foduucom/web-form-ui-field-detection model, then hints that you should specify the model and the confidence. Both `detect` and `find` take a confidence threshold, and the `config` tool exists partly so you can set a default, with 0.8 given as the example value. The README also warns in a hint that adding the words use imagesorcery makes the assistant pick the right tool, which is an admission that tool selection is not deterministic. Per tool detail lives in separate files under src/imagesorcery_mcp, one README.md each for tools, resources and prompts, so the main file is a summary rather than the reference.
pipx install, then a post-install step that writes the client config
The install is two commands, and the second one matters as much as the first:
pipx install imagesorcery-mcp
imagesorcery-mcp --post-installThe first puts the package in its own environment under pipx, which the README calls its recommended installer for virtual environment management. The second is a console entry point declared in pyproject.toml as imagesorcery-mcp = "imagesorcery_mcp:main", and the visible portion of the Cline walkthrough puts it at step three, between loading the MCP documentation and editing the MCP-configuration file. That ordering tells you the post-install step does not create your client configuration for you; you still have to open that file and add the server entry yourself. Four more console scripts sit next to it in pyproject.toml: download-yolo-models, download-clip-models, create-model-descriptions and post-install-imagesorcery, whose target path is cut off mid word in the packaged file. The presence of separate model download commands, with requests, tqdm and huggingface_hub among the dependencies, is the reason first use is not instant.
Four OpenCV system libraries are the real install friction
Python 3.10 or higher is the only stated version requirement. The rest is system packages, and the documentation is candid that they arrive with an OpenCV install but go missing inside slim images. On Ubuntu and Debian:
sudo apt-get update && sudo apt-get install -y ffmpeg libsm6 libxext6 libgl1-mesa-glxFor containers the same four packages go into a Dockerfile instead:
RUN apt-get update && apt-get install -y ffmpeg libsm6 libxext6 libgl1-mesa-glxThat is the whole dependency story stated in the README, and it is the step most likely to fail silently in CI, where a base image has none of them. The client side is equally short: Claude.app, Cline, or another MCP client. Note what is not promised anywhere: no GPU requirement, no model size, no memory figure, and no statement about which platforms beyond Ubuntu and Debian are supported.
Two telemetry libraries sit inside a local only package
The capability list ends with a claim that all of the work happens locally, without sending your images to any servers, and every tool description supports that reading: nothing in the table takes a URL as input. The dependency list complicates it. pyproject.toml includes amplitude-analytics and posthog, each carrying the comment for telemetry, alongside python-dotenv for loading environment variables from a .env file. The README does not say what those two libraries collect, whether they can be switched off, or how to disable them, and it does not document a single configuration key for doing so. There is a CONFIG.md file at the top level of the repository and a `config` tool that views and updates configuration settings, so the settings surface exists. Whether it covers telemetry is something you have to open the file to find out. If a no upload guarantee is a requirement rather than a preference, that file is the thing to read before you install.
The install walkthrough stops at step five, and LLM-INSTALL.md is the rest
The README ends mid sentence inside the Cline section, at step five of the numbered install list, with the line Add the imagesorcery-mcp s and nothing after it. Everything past that point is not in the file you are reading, so treat any claim about the full client configuration as unverified. The top level of the repository shows where the missing text went: LLM-INSTALL.md holds the copy paste install sequence for Cline, CONFIG.md covers configuration, GEMINI.md is a client specific companion, and setup.sh and pytest.ini sit beside src/ and tests/. The last push to the repository was on 2026-05-19, it is not archived, and it has no GitHub releases, so the version in pyproject.toml is the only version marker a reader can check.
Editorial conclusion
ImageSorcery MCP fits a workflow where an assistant edits files already on your disk and you do not want images uploaded anywhere. Before adopting it, confirm the four OpenCV system libraries exist in your container image, read what amplitude-analytics and posthog report before you repeat the local only claim to anyone, and pin the version yourself: pyproject.toml says 0.12.0 and there are no GitHub releases to roll back to.
Frequently asked questions
Which software is best for image processing?
It depends on the operation rather than the tool. ImageSorcery MCP splits the work across OpenCV for geometry, Ultralytics for object detection, EasyOCR for text extraction and Pillow for metadata, and reaches them through an MCP client such as Claude.app or Cline.
What are the different methods used in image processing?
The methods are per tool in this server. crop uses OpenCV's NumPy slicing, rotate calls the imutils.rotate_bound function, blur and fill take rectangular, polygonal or mask based areas and can invert them, detect and find return segmentation masks as PNG files or polygons, and ocr runs EasyOCR.
Does ImageSorcery MCP send my images to a server?
The capability list says everything runs locally, without sending your images to any servers, and no tool in the table accepts a URL. The pyproject.toml dependency list also carries amplitude-analytics and posthog, both commented for telemetry, and the README does not say what they report.
What can the config tool in ImageSorcery MCP change?
It views and updates ImageSorcery MCP configuration settings, with setting the default detection confidence to 0.8 given as the example. A CONFIG.md file sits at the top level of the repository, and python-dotenv is a dependency for loading environment variables from a .env file.
How do I install ImageSorcery MCP inside a Docker container?
The README says the OpenCV system libraries can be absent from virtual environments such as Docker, and gives one Dockerfile line installing ffmpeg, libsm6, libxext6 and libgl1-mesa-glx. The package itself installs with pipx install imagesorcery-mcp and needs Python 3.10 or higher.
Official sources
Add this badge to your README
If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.
[](https://hysenlabs.com/projects/sunriseapps-imagesorcery-mcp)Star history chart for your README
[](https://hysenlabs.com/projects/sunriseapps-imagesorcery-mcp)Was this useful?