
Remove AI Watermarks MCP Server: Setup for Claude, Cursor and Other Apps
Lets Claude and Cursor remove AI watermarks and provenance metadata from your own images and video, or check which marks a file carries.
This is the Remove AI Watermarks project's own MCP server, started with the mcp subcommand of its command-line tool. The tool strips AI provenance marks: visible labels such as the Gemini sparkle, C2PA, EXIF, XMP and IPTC metadata, and, for images on a CUDA GPU, invisible pixel watermarks. The server has three tools that check a file, clean one file, or clean a whole folder.
It runs on your computer and reads and writes files there. The project says it is for content you made or edited yourself, not for stock or agency marks that protect someone else's work. Some modes need extra packages, and the invisible-mark scrub needs a CUDA GPU.
Good for
- Artists who want to strip metadata from images they generated
- Creators cleaning a folder of their own renders in one request
- Anyone who wants to see which AI marks a file carries
Works in
12 should work
Tap an app to see how to set it up and any known problems.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Client does not support this transport.
Client does not support this transport.
Install Remove AI Watermarks MCP
Claude Code
- 1
Copy this prompt.
Add the Remove AI Watermarks MCP server to Claude Code. Run this command: claude mcp add remove-ai-watermarks -- uvx remove-ai-watermarks mcp Then check that it loaded and tell me if I need to sign in or restart.
- 2
Paste it into Claude Code. It shows the change and asks before running it.
Requires uv.
For visible-mark removal and video, name the extras: uvx --from "remove-ai-watermarks[visible,video]" remove-ai-watermarks mcp. Without them, the tools return an error with the install hint.
Built from the package data we hold. Not yet tested by us. Source.
Tools
Checks an image or video for AI watermarks and provenance data
Reports which AI watermarks and provenance metadata a file carries. It can skip the visible-mark check. It changes nothing.
Try: “Check if photo.png carries AI provenance data.”
Runs code ·
identifyRemoves AI watermarks and metadata from one file
Writes a cleaned copy next to the original, named <name>_clean, or to the path you give. The original is never overwritten. The all mode also tries to scrub invisible marks in images when extras and a CUDA GPU allow it.
Try: “Remove the metadata from my-art.png.”
Runs code ·
removeCleans every file in a folder
Cleans a whole folder into a new output folder named <dir>_clean, or the one you give. It refuses to write over the input folder.
Try: “Clean the metadata on every image in ./renders.”
Runs code ·
batch
Read from the source code
Permissions
medium risk · 3 things to watch
Medium risk
It reads files you point it at and writes cleaned copies beside them. It never overwrites the original, so check the output path before you ask it to clean a folder.
What it can do
- Read your data
- Create and change things
- Delete things
- Read-only mode
Watch out for
It writes new files on your disk
The remove and batch tools create cleaned copies in a _clean file or folder. Check the output location. github.com
It runs the command-line tool as a child process
Each tool runs one CLI command and returns its output. Only point it at files you own. github.com
Intended for your own content
The project states it is for lawful use on content you own, provided as is, and does not target stock agency previews or other marks that protect third-party paid content. Its README also says some jurisdictions, including China, restrict removing AI labels or supplying tools that do. github.com
Security advisories
No published security advisories as of Nov 1, 2026.
Technical details
| Sign-in and keys | No sign-in or key. |
|---|
Questions
What does Remove AI Watermarks remove?
It removes AI provenance marks from images and video: visible labels such as the Gemini sparkle, C2PA, EXIF, XMP and IPTC metadata, and, for images with a CUDA GPU and extras installed, invisible pixel watermarks. Through the MCP server, metadata mode is the default; visible and all modes are chosen per call.
Is the Remove AI Watermarks MCP server official?
Yes. It ships inside the Remove AI Watermarks project itself, as the mcp subcommand, and is listed in the official MCP registry.
Does it need an API key?
No. It runs locally with no sign-in or key.
Can it overwrite my original files?
No. The docs say the server refuses an output path equal to the source, and an output folder equal to the input folder.
Is it local or hosted?
Local. Paths in tool calls are read and written on the computer that runs the server.
What can I use it on?
The project says it is for content you generated or edited yourself, not for stock, agency or marketplace marks that protect a third party's asset.
Is removing AI labels legal?
It depends on where you are. The README says some jurisdictions, including China, restrict removing AI labels or supplying tools that do, and that each user is responsible for lawful use. It is not legal advice.
Information
updated Nov 2, 2026
- Publisher
- Official
- Package
remove-ai-watermarks- Registry
- PyPI
- Transport
- stdio
- Auth
- None
- Version
- 0.52.0
- License
- Apache-2.0
- Last commit
- Oct 30, 2026
- Open issues
- 2
- Listed in
- Official MCP registry
Sources
Docs, issue trackers and discussions this page is built from. Issues, quotes and findings above link to their own source.
Show all 2Hide
- github.com · github.com
- github.com · github.com