You can now ask your AI assistant to clean files for you: "remove the location from the videos in my Downloads folder", or "strip my name from this PDF before I send it". This works through MCP (Model Context Protocol), the standard way assistants like Claude, Cursor and VS Code use outside tools. This post shows how to set up our metadata remover MCP server, what a real session looks like, and what the assistant can and can't see.

Can Claude, Cursor or ChatGPT remove metadata from files?

Our MCP server gives your assistant two tools:

  • clean_files: cleans files or folders on your computer and saves a clean copy next to each one, as name-clean.ext. Your originals are never changed.
  • check_usage: how much of this month's allowance is left.

It cleans the same files as our website and API:

Kind Formats What's removed
Video MP4, MOV, M4V, 3GP, MKV, WebM GPS and GPS tracks, phone or camera model, serial numbers, dates, author, hidden data tracks. Not re-encoded
PDF PDF Author, software, dates, hidden document data, metadata inside embedded photos, digital signatures (optional)
RAW CR2, CR3, NEF, ARW, RAF, RW2, DNG GPS, serial numbers, owner, dates, shutter count. The photo data is untouched
Photos JPEG, PNG, WebP Location, device, dates, Content Credentials, AI prompts and settings

Most EXIF MCP servers listed today read metadata, and a few strip it from photos. This one removes it from video, PDF and RAW files too, and checks every result.

What does it look like in practice?

We pointed it at a test folder, trip, holding an iPhone video, a Word PDF, a Canon RAW file and two photos. Asked to clean the folder, the assistant called clean_files, which returned:

Cleaned 4 of 5 files.

beach.mov: Clean
  saved to: trip/beach-clean.mov
  removed: Dates and times, Device make and model, GPS location, Software / OS version,
           Text fields (titles, captions, comments),
           Timed metadata tracks (e.g. face detection, scene data)
itinerary.pdf: Clean
  saved to: trip/itinerary-clean.pdf
  removed: Author / owner name, Dates and times, Software / OS version, Unique IDs, XMP metadata
sunset.cr3: Clean
  saved to: trip/sunset-clean.cr3
  removed: Author / owner name, Dates and times, Device serial number, GPS location, Unique IDs
  kept on purpose: Device make and model
  note: Camera make, model and lens are kept: raw editors need them for colour profiles
        and lens corrections.
IMG_2042.jpg: Clean
IMG_2041.jpg: Not cleaned
  We could not clean this file safely, so nothing was returned.

The assistant then tells you, in plain words, what was in each file and where the clean copies are. When a file can't be cleaned safely, it says so and returns nothing, rather than handing back a damaged or half-cleaned copy. The photo that wasn't cleaned exposed a bug on our side: our check was too strict with very vivid colours in the iPhone colour profile. That's fixed.

How to add the metadata remover to Claude, Cursor or VS Code

You need a Business plan API key (from the API keys page) and Node.js 18 or newer.

Claude Desktop: Settings → Developer → Edit Config, add this, put your key in, and restart:

{
  "mcpServers": {
    "aimetadatacleaner": {
      "command": "npx",
      "args": ["-y", "aimetadatacleaner-mcp"],
      "env": { "AMC_API_KEY": "amc_live_..." }
    }
  }
}

On Windows, if Claude can't start it, use "command": "cmd" and "args": ["/c", "npx", "-y", "aimetadatacleaner-mcp"].

Claude Code: run once:

claude mcp add aimetadatacleaner -e AMC_API_KEY=amc_live_... -- npx -y aimetadatacleaner-mcp

Cursor: Cursor Settings → MCP → Add new MCP server, or add the same block as Claude Desktop to ~/.cursor/mcp.json.

VS Code: create .vscode/mcp.json. VS Code asks for the key the first time and stores it safely:

{
  "inputs": [
    { "type": "promptString", "id": "amc-key", "description": "AI Metadata Cleaner API key", "password": true }
  ],
  "servers": {
    "aimetadatacleaner": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "aimetadatacleaner-mcp"],
      "env": { "AMC_API_KEY": "${input:amc-key}" }
    }
  }
}

Windsurf, Cline and most other MCP apps take the same block as Claude Desktop. Our MCP setup page has every option.

Does it work on claude.ai and in ChatGPT?

These run in the browser, so they connect to our server directly, and you sign in with your AI Metadata Cleaner account instead of using a key:

  • claude.ai: Settings → Connectors → Add custom connector, name it AI Metadata Cleaner, and paste https://api.aimetadatacleaner.com/mcp. Click Connect, sign in and choose Allow.
  • ChatGPT: turn on developer mode in Settings → Apps → Advanced settings, then create an app with the same URL and OAuth sign-in.

One limit: these apps can't pass a file you drop into the chat on to a connector. There, it cleans files that are online at a direct link (a Dropbox download link, for example). For files on your computer, use Claude Desktop, Claude Code, Cursor or VS Code.

Does the AI see my files?

The add-on sends each file from your computer straight to our server and saves the clean copy on your computer. The assistant only sees the file names and the report of what was removed, never the contents. Files are cleaned on our server in Germany. Each upload is deleted as soon as cleaning finishes, and the clean copy once it's saved back, or after 15 minutes. We record the file type, size and result for billing, never the name or contents.

Why not just ask the assistant to write a script?

Claude Code or Cursor can write you an ExifTool or ffmpeg command in seconds. The trouble is that the usual commands fail silently:

  • ExifTool's PDF edits can be undone.
  • A blunt ExifTool strip can change how a RAW photo looks.
  • The common ffmpeg shortcut -map 0 copies iPhone's hidden face-detection tracks.
  • Unquoted ffmpeg options fail outright in the Mac's default shell.

We tested these in what ffmpeg and ExifTool miss. A generated script exits happily either way. The MCP server runs checked methods and re-reads every result, so "clean" means it was checked.

How much does it cost?

MCP is part of the Business plan: $49 a month, or $449 a year. That's 10,000 files and 1TB a month, files up to 2GB, shared with the API. Files that fail or aren't supported don't count. A folder means the supported files directly inside it, not the folders inside those.

Frequently asked questions

Is there an MCP server that removes EXIF data?

Yes. Ours removes EXIF, GPS, XMP, Content Credentials and AI generation data from photos, and the equivalent details from video, PDF and RAW files. It works in Claude Desktop, Claude Code, Cursor, VS Code, Windsurf and on claude.ai.

Are my original files changed?

Never. Clean copies are saved next to the originals with "-clean" added to the name.

Do I need to code?

No. You paste one block of settings into your AI app once. After that you just ask.

Can I clean a whole folder?

Yes. Name the folder and it cleans the supported files directly inside it. Business covers 10,000 files a month.