Let your AI assistant clean images.

What the MCP server can do

Clean one image with clean_image

Give your assistant the absolute path to a JPEG, PNG, WebP, or AVIF file and ask it to make a metadata-clean copy. The tool checks that the path points to a real supported image, enforces safe file and pixel limits, creates a fresh JPEG, verifies the result, and returns the new absolute path. The source image is never overwritten. If a file with the planned output name already exists, the server chooses another safe name rather than replacing your work.

A useful request is: “Clean the metadata from /Users/me/Pictures/example.png and keep the original.” Your assistant can call the tool and tell you exactly where the cleaned copy was written. Paths must be absolute so the action is explicit and predictable; ambiguous relative paths are rejected.

Clean a batch with clean_images

The batch tool accepts up to 50 absolute image paths in one call. You can also provide an absolute output directory and a JPEG quality from 40 through 100. Each image is handled separately, so the structured result explains which files succeeded and which failed. That makes batch work easier to review without hiding a damaged, oversized, missing, or unsupported input among successful results.

Try asking: “Use aimetadataremover to clean these product photos and place the new copies in /Users/me/Desktop/clean-photos.” The assistant still needs access to those paths under its own permission model. The MCP server does not expand access beyond what your local assistant already has.

What local means

The assistant starts the server as a local child process and communicates through standard input and output. The image bytes are read from the path you approve and the new file is written locally. Only the text result, such as the output path, dimensions, and status, returns to the assistant. Your chosen AI client may have its own data and permission policies, so review those separately. aimetadataremover.co cannot control how a third-party assistant handles your conversation.

Supported formats and safety limits

Inputs can be JPEG, PNG, WebP, or AVIF, up to 10 MB each and up to 30 million pixels. Outputs are clean JPEG copies. A batch can contain no more than 50 images. Quality must be a whole number from 40 to 100. These boundaries keep tool calls predictable and reduce the chance that a malformed or unexpectedly huge image exhausts the computer’s memory.

The package avoids destructive behavior. It rejects non-files, unsupported extensions, relative paths, oversized files, oversized dimensions, and unsafe quality settings. It writes through a temporary file and only publishes a completed result. Temporary artifacts are removed when an operation fails. The proprietary engine is distributed as a production package without readable source maps; its license forbids redistribution, reverse engineer activity, resale, or creation of a competing clone where those restrictions are lawful.

How to confirm it is connected

Restart the client after adding the server if its tool list does not refresh automatically. Then ask the assistant to list the aimetadataremover tools. You should see clean_image and clean_images. Start with a copy of a non-sensitive picture and inspect the returned output path. Keep original masters in a private archive whenever capture dates, GPS records, author fields, color profiles, or authenticity information may be useful later.

Important scope

This tool removes common hidden file metadata from the new sharing copy. It does not erase text, faces, landmarks, logos, visible watermarks, or labels baked into the picture. It also cannot control whether an external platform applies an AI label using its own classifiers or records from a previous upload. Use the MCP server for privacy-focused file sanitization, not as a promise about another service’s moderation or authenticity decisions.