From prompt to parcel: an MCP server that prints what you make in ChatGPT and Claude

Image models got good enough that people want their results on the wall, on a mug or on a shirt. The usual route is clumsy: download the image (often a compressed copy), find a print shop, guess the right size, upload, hope the crop works out.

We built MyImage2Life to shorten that to one conversation. It is a remote MCP server, so ChatGPT and Claude can call it directly. You describe what you want, the assistant makes or picks the design, you see a mockup made from the real print file, a price for your country, and after payment a print shop produces and ships it.

The tool flow

The server exposes a small set of tools that map to the steps a person would take anyway:

• search_products: posters, canvas, wood, aluminium, acrylic, mugs, T-shirts, cards, calendars, puzzles, cushions and more
• create_design or an upload of your own photo
• create_proof: a mockup rendered from the actual print file, with an honest warning when the resolution is too low for the chosen size
• get_price: price including shipping for the delivery country
• place_order: returns a payment link; nothing is printed before payment
• get_order_status

The endpoint is https://myimage2life.com/mcp, and the server is listed in the official MCP registry as com.myimage2life/print. This page explains how to connect it in ChatGPT and Claude.

Lessons from building it

1. Assistants read text, not your UI. A tool result can carry a nice card, but the model reasons over the text part. Every id the next tool needs (design id, proof id) has to be in that text, or the assistant starts guessing.

2. Clients cache the tool list. After changing tool schemas we had to disconnect and reconnect the connector and start a new conversation; a running conversation keeps the old schema.

3. Print size is the real quality problem. Most generated images are 1024 to 1536 px on the long side. At 300 ppi that is 9 to 13 cm. We wrote up the numbers in how big can you print an AI image, and there is a free print size checker that does the math per format.

4. The crop belongs to the product. A mug print area is a wide strip, a poster is 2:3. Generating or editing in the shape of the print area beats cropping afterwards; otherwise most of a square image can disappear on a mug.

5. Blur has three usual causes: a file too small for the format, compression on the way, or soft areas a phone screen hides. More in why AI images look blurry in print.

If you are building an MCP server that ends in a real-world transaction, I am happy to compare notes in the comments.