Yes, by redrawing them. ChatGPT's image tool changes a photo by generating a new one. For exact edits like resize, crop and compress, an Image server keeps the real pixels.
ChatGPT edits photos by redrawing them. The Image server makes exact edits on the real pixels, resize, crop, compress, convert and strip metadata. It runs hosted or locally in Claude and Cursor.
What you get
Product photos for the shop
Etsy seller of jewellery in Gdansk
| Tool call | What it did |
|---|---|
| image_batch_resize | Resizes the 5 photos |
| image_strip_metadata | Removes GPS and camera data from each copy |
5 web-ready photos with no location data, and the originals unchanged.
Free to start. Image Pro is $19 once. Free vs Pro
ChatGPT's image tool makes a new picture from your photo, so faces and small details can change. Its code tool can resize or crop an upload, one chat at a time.
The Image server resizes, crops, compresses, converts format, makes thumbnails and strips metadata, always writing a new copy. Watermarks need a local install, which ChatGPT can't reach. Free sources go up to 4 MP. The hosted server can't see files on your computer. It takes a file from a public link or a small upload (up to 2 MB). To try it in ChatGPT on the web, where your plan or workspace allows custom MCP servers, get a free token at mcp.zovo.one/mcp/connect and add mcp.zovo.one/mcp/image/t/<token> as a server with No authentication. We haven't tested this in ChatGPT yet.
Example scenarios built on the real tools of Image. Names and amounts are invented.
Resize product.jpg to 1200 pixels wide, then compress it for the web.
A web-sized product photo with the same pixels, just smaller, and the original untouched.
Resize these 5 photos to 1200 px wide and remove the location data before I list them.
5 web-ready photos with no location data, and the originals unchanged.
About 15 min saved (estimate)
Compress hero.png under 300 KB and convert it to JPEG.
A JPEG under 300 KB with the before and after byte counts.
About 5 min saved (estimate)
Every example here works on the free tier. One $39 key removes the limits on all of them. Pricing
Paste one URL into Claude.ai, Claude Desktop, Cursor or VS Code. No install, no account.
1. Get a free token (30 days, 600 calls an hour)
curl -s -X POST https://mcp.zovo.one/mcp/token
2. Paste this URL as a custom connector
https://mcp.zovo.one/mcp/image/t/<token>
Or run it locally (Node.js 18+)
npx -y @theluckystrike/mcp-image
| Free ($0, no key) | Pro ($19 once) |
|---|---|
| Outputs up to 4 MP | Any output size |
| Batch and thumbnails up to 5 files per call | Any number of files per call |
| Watermark with your business name (local install) | Any watermark text (local install) |
| Top 3 dominant colours | Up to 16 dominant colours |
| image_info and strip metadata at any size |
servers/image/src/index.ts:24-28 (FREE_MAX_PIXELS = 4_000_000, FREE_MAX_BATCH = 5, FREE_MAX_COLORS = 3), :509 (custom watermark)
Yes. Claude edits uploaded .xlsx and CSV files with code execution, on by default on Free, Pro and Max. When the Spreadsheet server still helps.
You get: Table from your file PDFYes. Claude merges and splits PDFs you upload, with code execution on by default for Free, Pro and Max. The PDF server covers files on disk and Cursor.
You get: Stamped PDF QuotesLine items with VAT, a validity date and a PDF, made from one prompt. Turn the accepted quote into an invoice. Uses the Quotes server. Free tier, Pro $19 once.
You get: QuoteImage Pro, $19 once
Analytics cookies
We use Google Analytics to count page views and clicks on buy links. In the EEA, the UK and Switzerland it sets cookies only if you accept. No ads, no sale of data. Privacy Policy