Pure scratch-my-own-itch. I’ll dive deep into the cloud code, get the app basically ready, then I’ll need a product shot, an icon, an OG image, so I’ll go to MidJourney, a photo studio, or Figma, explain my brand again for the fifth time, and get something back. The app worked fine. It looked as if a robot had made it, which is fair enough, someone had.
We tried everything (v0, Figma, Base44, cloud design tools), and they all hit the same level: emotional, on-brand identity (think Duolingo, think Fantom) that separates a mediocre site from a great one, and that’s exactly what AI builders can’t do.
So I gave my coding agent a paintbrush. AgentBrush is an MCP server: your agent creates images inline, places them on brand via reference images, edits areas with the mask editor, removes backgrounds locally for free, and writes the file directly to your repo.
It is paid (from $6.99/month, every generation is a real GPU call) and it is v1. Nice to hear about Cassie, and I’m really curious what you’d like to say first.
<a href