Most developers I know have the same problem: the logic is solid, but the UI looks like it was built in a hurry — because it was.
I spent two days on a login flow for a side project before I gave up and tried a different approach: Google Stitch for the UI, Antigravity as the IDE, and the MCP bridge between them. Here's what the workflow actually looks like.
The core idea
Google Stitch gives you pre-configured UI patterns — "Stitches" — that are already accessible and mathematically sound. You pick a functional category (Hero, Card, List), set one primary color and one font, and it handles the rest. The MCP connection to Antigravity means you never manually export manifests: change a button style in the Stitch web UI, and it reflects in your running simulator instantly.
The implementation
Once you've connected Stitch to Antigravity via MCP: Configure Server in the command palette, your components become requestable by name:
import { useStitch } from '@antigravity/react-hooks';
function DashboardCard({ userId }) {
const { Component, loading } = useStitch('UserCard');
if (loading) return <Placeholder />;
return (
<Component
username={userId.name}
onAction={(ev) => console.log('User tapped:', ev)}
/>
);
}
Your React component has no idea what UserCard looks like. It just requests it. If you decide a List works better than a Card, you change it in Stitch — the code stays the same.
What I didn't expect
The MCP connection is bidirectional. You can send performance feedback from Antigravity back to Stitch, so the design tool knows which components are causing frame drops on real devices. The documentation barely mentions this — it took me a while to find it.
The full guide
I wrote a more detailed walkthrough on Medium covering the full setup, the Gravity Fields fetch pattern, and the telemetry configuration: https://medium.com/p/02aa7c97e131
What's your approach for UI as a solo dev — do you start from the design or the logic?
Top comments (0)