Refero
Design research for humans and AI
4.9•15 reviews•1.6K followers
Design research for humans and AI
4.9•15 reviews•1.6K followers
Real product screens, flows, and patterns to help teams and AI make better design decisions.
This is the 5th launch from Refero. View more
Refero MCP
Launched this week
Refero MCP connects your agent to a curated library of real product interfaces and user flows. It studies before it builds — and the output looks designed, not generated.
Free Options
Launch Team
Refero
Hey Product Hunt, I’m Mike, founder of Refero.
We built Refero MCP because most AI-generated interfaces still look generic. Models are great at code and logic, but product design is a different skill. They usually don’t know which patterns work, how real products structure flows, or what makes an interface feel thoughtful.
Refero MCP gives AI agents access to 125,000+ real product screens and 8,000+ user flows, so instead of designing in a vacuum, they can study real products before generating UI.
The idea is simple: better references lead to better design output.
Less guessing, less generic UI, and more interfaces that actually feel designed.
Try it live: https://demo.refero.design/ or integrate into your agent https://doc.refero.design/mcp/getting-started
Would love your feedback <3
Raycast
@mishkadoing this is pretty slick — but the reasoning step took a long time! Why is your inference so slow?
Refero
@chrismessina Thanks for trying it out (legend 🙌). In the demo we actually let the agent overthink a bit. It calls a lot of tools and does a bunch of research before answering. The MCP server itself is fast. Works perfect inside Claude Code or Codex.
Tried this with a "node-based editor" prompt and the results were solid
couple of small things:
- explore next suggestions might be buttons so i don't have to copy-paste it
- downloadable report would be very useful. like a PDF or .md report that i can pass to the agents.
- to be able to pass my existing UI or website and get suggestions would be cool
Great tool, planning to use it later again!
Pinnacle
Interesting approach using MCP to inject design context how are you handling auth on the server side? Curious whether clients authenticate per-request or whether there's a session/token model, since that changes a lot about how agents can be scoped to specific design systems.
So cool! Was looking for this when writing a report, does it work for agents that help in ppt/doc design?
AutonomyAI
Played around with the demo and it is super useful. If only for UI research. Very cool! shared it with our PM. Best of luck!