project //
Figma MCP Plugin - AI-Assisted Layout Generation
This project is actively in development. What follows reflects the thinking behind it and where it stands today
Overview
Design systems solve the consistency problem. They don’t solve the speed problem.
Even with a well-structured component library, moving from a brief to a first layout exploration takes time. Time spent on decisions that, increasingly, don’t need to take that long.
While rebuilding Ellipse into a more modular system, I started wondering what it would look like to connect that component architecture directly to an AI generation layer. If the components were truly composable and the tokens were structured correctly, could a prompt generate a working page layout that already lived inside the system?
That question turned into a Figma MCP plugin.
The Opportunity
We were in the middle of a design system migration, breaking components into smaller, more composable building blocks. The timing created an interesting opening.
If we built the new system with AI generation in mind from the start, the plugin could take advantage of that modularity immediately. The modular rebuild wasn’t just about flexibility for designers. It was about building a system that an AI layer could actually reason about.
How It Works
The plugin connects Figma to an MCP layer that understands the Ellipse component library, token structure, and layout patterns. A designer pastes a prompt describing a page or section. The plugin generates layouts across breakpoints, stacks components, and places them on the canvas following the system’s existing rules.
It’s not generating random designs. It’s working within the system we already built.
Early results show it generating multi-breakpoint page layouts that respect token values and component composition patterns, dramatically reducing the time from brief to first explorations.
What Made It Possible
The Ellipse rebuild was the prerequisite. Rigid, single-use components can’t be composed intelligently by an AI layer. They were built for humans who understand the exceptions and the workarounds. The new architecture, built around smaller reusable building blocks, gave the plugin a foundation it could actually work with.
The quality of what gets generated is only as good as the quality of the system underneath it. That part doesn’t get talked about much in conversations about AI design tools.
Early Results
This project is actively in development. What follows reflects where things stand today.
• Generates multi-breakpoint page layouts from a single prompt
• Components placed on canvas follow Ellipse token and composition rules
• Reduces time from brief to first layout explorations
• Being developed in parallel with the ongoing Ellipse component migration
What's Next
As the component migration completes, the plugin’s range expands. More components means more layout patterns and more complex page architectures from a single prompt.
The goal over time is a design team that uses AI to handle the scaffolding work, so designers can focus on the decisions that actually require judgment and taste.