# Demarker [Health: Active]

**Category:** 🎥 Multimedia Process  
**Repository:** https://github.com/removeaiwatermark/demarker  
**GitHub Stars:** 0  
**npm Downloads (last month):** 7  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/demarker

## Description
Identify what an AI image carries (marks, metadata, watermarks), then clean it.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "demarker": {
    "command": "npx",
    "args": ["-y","demarker"],
    "env": {
      "DEMARKER_API_KEY": ""
    }
  }
}
```

**Requires environment variables:** `DEMARKER_API_KEY` — the values above are empty placeholders; fill in real credentials before running (see the repository for what each one is for).

## Documentation

## What the Demarker MCP server does

The Demarker MCP server exposes Demarker’s image identification and cleaning workflow to MCP-compatible clients. Agents can use it to examine an image for visible marks, embedded AI metadata, and inferred hidden watermarks, including signals associated with SynthID. They can then request cleaning when the user has permission to do so.

Identification does not require an account and is free according to the project documentation. Cleaning is a separate paid operation: it requires an API key, available credits, and an explicit confirmation of the relevant rights statement. The server is intended for use from agent clients such as Claude, Cursor, and VS Code.

## How it works

An agent sends an image operation through MCP. Identification analyzes the image and reports what Demarker detects. Cleaning submits the image for removal of the detected signals and returns a result without replacing an existing file.

The workflow includes safeguards around paid actions. A cleaning request must include a credit limit, and the MCP server applies its own per-call ceiling, which defaults to two credits. Rights confirmation is not assumed. Where the client supports prompting, the server asks the user for that confirmation.

## Setup and configuration

The repository provides a direct MCP launch command:

```bash
claude mcp add demarker -- npx -y @demarker/mcp
```

The package can therefore be started through `npx` without a separate build step for normal use. One-click installation options for Cursor and VS Code are also linked from the project’s developer page.

Set `DEMARKER_API_KEY` before using cleaning operations. The key is obtained from the API settings area at Demarker. No key is needed for scanning. The CLI examples show the same service through commands such as `npx demarker identify image.png` and `npx demarker clean image.png --max-credits 1 --attest-rights`.

## Tools and capabilities

The Demarker MCP server supports these documented operations:

- Identify visible marks in an image.
- Inspect AI-related metadata.
- Infer hidden watermark signals, such as SynthID.
- Clean an image after authorization and credit controls are supplied.
- Return results without overwriting an existing file.

The repository also contains a command-line package, a typed client, and an agent skill, but those are separate packages rather than additional MCP tools described in the excerpt.

## Limitations and notes

Scanning and cleaning have different access requirements. Scanning is free and does not need an account, while cleaning consumes credits and needs an API key. Each cleaning request also requires the user to confirm a rights statement; the CLI requires an explicit `--attest-rights` option, and the MCP server asks through the client where that behavior is supported.

Credit spending is bounded, but callers still need to provide a maximum. The server’s default per-call ceiling is two credits. Existing files are not overwritten by results. The documentation does not specify the full set of image formats, the exact detection response schema, or guaranteed detection for every hidden watermark.

_Full upstream README: https://allmcps.com/mcp/demarker/readme_