Design Cortex: Turning a Figma System Into Something an AI Agent Can Actually Read
Design Cortex started as a fix for my own workflow: every time an AI agent needed to know something about a design system, it queried Figma live, burning tokens and time on the same lookups over and over. It's now open source, MIT licensed, and works with any Figma MCP. It also just launched on Product Hunt, so here's what it actually does and why I built it the way I did.
The problem: querying Figma live doesn't scale
Most AI design workflows have an agent hit Figma directly on every task: pull the component, read its variants, check the tokens, repeat. That's slow, it's expensive at scale, and a single request can pull far more data than the agent actually needs for the task in front of it.
The fix isn't a smarter query, it's decoupling extraction from consumption. Pull the system once, structure it well, and let agents reference it locally for as long as it stays fresh. That's the whole premise behind Design Cortex: a one-time extraction into a markdown knowledge base, so the agent reads local files instead of hitting Figma mid-task.
Four skills, one job each
The pipeline is four small, composable skills instead of one big script:
ds-extract crawls Figma, caches the raw data, and produces a gap report.
ds-write converts that cache into a full markdown knowledge base.
ds-refresh re-extracts and updates only the components that changed.
ds-validate checks completeness, references, and freshness.
Each one fires from a plain-language trigger, and configuration lives in a single .ds-kb-config.json file, so the whole pipeline is reproducible on any machine, not just the one it was set up on.
A knowledge base built for token budgets
Agents load a top-level index.json table of contents, then a component index.md, then only the specific variant files they actually need. That layering is what keeps token usage flat no matter how big the design system gets.
Figma stays the source of truth throughout. The knowledge base is a derived artifact, regenerated on refresh, never a second place to manually maintain. It's MCP-agnostic and dependency-free, so it drops into Cursor, Claude Code, or anything else that speaks MCP, with no API keys, no Python, and no services to stand up.
Featured on Product Hunt
Design Cortex launched on Product Hunt shortly after I open sourced it, which was a good forcing function to make the README, the setup flow, and the first-run experience actually usable by someone who isn't me. The response confirmed the core bet: a lot of people are wiring AI agents into design systems right now, and most of them are hitting the same live-query wall.
Try it
Design Cortex is MIT licensed and available on GitHub. Point it at your Figma file, run the extraction once, and your agent gets a fast, local source of truth instead of a live dependency on Figma's API. If you try it and hit a rough edge, open an issue. It's still early, and the workflow gets better with real usage.