SkillFoxx

watermarkless (zigansh1n) MCP

watermarkless

Watermark cleaning for photos for an agent: Avito and Cian profiles, automatic mark search and a custom mask, LaMa model run locally

  • 0 ★ on the repository
  • Last commit
MCP server

Broad access

Why this level

  • Removing watermarks from other people's photos infringes copyright and site terms, and the user is responsible
  • Cleaning competitors' or other people's listing photos can get an account blocked on the platform
  • Running the HTTP service with no token or with `CLEANER_ALLOW_ANON=1` opens file processing to the whole network
All reasons and checks
Russian stack

zigansh1n/watermarklessAuthor:zigansh1n

Install

Services

Manual install

claude mcp add -s user watermarkless -- /path/to/repo/.venv/bin/python /path/to/repo/mcp_server.py

After cloning the repository and installing `requirements.txt` into a Python 3.10 or newer environment.

Access and checks

How we review

Broad accessThe tool can change data, run commands or handle secrets.

  • Removing watermarks from other people's photos infringes copyright and site terms, and the user is responsible
  • Cleaning competitors' or other people's listing photos can get an account blocked on the platform
  • Running the HTTP service with no token or with `CLEANER_ALLOW_ANON=1` opens file processing to the whole network

This is third-party code. Review the repository files before installing.

What it does

A Python MCP server removes watermarks from photos with LaMa inpainting. Files are processed on your own machine, originals are not changed and the result goes to a `clean` subfolder. The same code is also available as the `app.py` HTTP service that an Avito listing manager calls. Server tools: `list_profiles` shows known marks, `detect_watermark` estimates whether a mark is on a photo, `find_watermarks` finds boxes with a detector for any site, `clean_photo` cleans one file, `clean_folder` processes a folder of a downloaded listing. The `avito` and `cian` profiles are built in. The `auto` mode finds marks of other sites with a YOLO11x model, and `mask_box` paints over a given rectangle with no check. A custom profile is built with `fit_profile.py` from 40 to 200 samples. The server runs over stdio and a Docker build is provided. No license is stated in the repository.

Who it is for. Those who work with photos of their own or licensed Avito and Cian listings and want an agent to remove the site logo from a folder of images.

Good fit when

  • The Avito logo has to be removed from every photo in a downloaded listing folder
  • Before cleaning, the agent checks with `detect_watermark` whether a mark is present and leaves clean photos alone
  • A mark of another site is found in `auto` mode and the box is refined with `find_watermarks`
  • A given rectangle is painted over with `mask_box` with no check
  • For a site with no ready profile, a custom one is built from samples with `fit_profile.py`

Not a fit when

  • The photos belong to others and there are no rights to use them: removing a mark infringes their rights
  • A guaranteed clean result is required: traces remain on bright backgrounds and with changing marks
  • An Apple Silicon machine counting on GPU: the TorchScript model does not load on MPS, the CPU is used

Example request

Check the listing photo folder, remove the Avito logo where it is present and put the result into the clean subfolder.

MCP

Transport
stdio
Authentication
not required

Limitations

Python 3.10 or newer, the `mcp` package, PyTorch and the model weights are required: the Docker image is about 2.3 GB and the `auto` detector downloads 114 MB on first run. The last commit is from September 2026, the repository is small and has no license.

The Cian mark contains the listing number, so it has no presence check: the profile cleans every photo by its mask. The `unblend` mode is experimental and leaves coloured noise. Limits: a photo up to 25 MB and 4096 pixels on a side.

How to disable. Run `claude mcp remove watermarkless`, stop the container with `docker rm -f watermark-cleaner` if needed and delete the cloned directory.

FAQ

How do I connect watermarkless to Claude Code?

Install the dependencies into a virtual environment and run `claude mcp add -s user watermarkless -- <path>/.venv/bin/python <path>/mcp_server.py`. The transport is stdio, no separate port is needed.

Which sites are supported out of the box?

Avito and Cian: they have ready profiles. Marks of other sites are found by the `auto` mode, and a custom profile is built with `fit_profile.py`.

Are the source files changed?

No. The result goes to `<folder>/clean/<name>.jpg` unless `output_path` is set, and the originals stay untouched.

What does the presence check do?

`detect_watermark` computes a mark score for the photo corner. If there is no mark, `clean_photo` returns `no_watermark` and writes nothing, and `force=true` cleans anyway.

Is removing watermarks legal?

It depends on the rights to the photo. Your own photos and photos with permission can be processed, others cannot: a mark often protects the rights of the author or the site.

Discussion

No questions or reviews yet.

All Avito tools
Russian stack

Batch photo variation for Avito, Cian and marketplaces: several edited copies, fingerprint comparison and a command-line mode

CLIBroad access
Russian stack

An MCP server for Avito through a running Chrome over CDP: search, listings, photos, messenger, Delivery orders and closing your own listings

MCP serverBroad access
Russian stack

A Python MCP server on Playwright that lets an agent work on Avito from your account: search, item pages, messages to sellers and publishing listings

MCP serverBroad access6
Russian stack

A plugin with skills and distributed MCP servers for parsing the Avito catalog, price monitoring and notifications, with an honest legal risk section

PluginBroad access3
Russian stack

A remote MCP server with a single Avito search tool that returns only listings new since the last run

MCP serverExtended access
Russian stack

A compact MCP for Avito Messenger and items with user_id auto-detection, SSL bypass for corporate proxies, and an interactive deploy.sh

MCP serverBroad access

README in short

The README describes running via Docker and Python, the MCP server tools, the HTTP API with response codes, how profiles work, the `auto` and `mask_box` modes, cleaning methods and detector accuracy measurements.

README badge

Report a problem