Refero MCP: Let AI Do Design Research Before Drawing UI

2026-03-16 | Product Hunt | Official Site | 123 Votes

30-Second Quick Judgment

What is this?: It connects your AI Agent (Cursor, Claude Code, etc.) to a design reference library of 120,000 real product screenshots. It lets the AI "see" what real products look like before generating UI, so the output is no longer generic.

Is it worth watching?: Yes. If you use AI to write frontend code or design UI, this tool precisely solves the "AI interfaces all look the same" pain point. At $12/month, it's affordable and currently the only MCP service doing this.

Three Questions for You

Is it for me?

Target Users:

• Developers using Cursor/Claude Code/Goose to write frontend code
• Product Managers who need AI help with UI design
• Vibe Coding enthusiasts — those who want AI to generate interfaces but hate how they look

Am I the target? If you often tell AI that its UI is "too generic," "looks like a template," or "lacks design sense," then yes.

When would I use it?:

• Writing a new page in Cursor and wanting the AI to see how Notion or Airbnb did it first
• Building a SaaS product and needing to reference competitor onboarding/pricing/dashboard designs
• Vibe Coding when you want the result to look like a "real product" rather than a Bootstrap template

Is it useful?

ROI Judgment: For those who frequently use AI for UI, $12/month is a steal. It pays for itself by saving you just one manual search session. If you only do UI once a month, the 3-day free trial might be enough.

Is it delightful?

The Highlights:

• Plug and Play: Once MCP is set up, the AI automatically searches references before generating. No workflow changes needed.
• The "Aha!" Moment: Watching AI output transform from "templated" to "looking like a real product."
• 120,000 Real Screenshots: These aren't flashy Dribbble concepts; they are real, live products like Figma, Shopify, and Airbnb.

The "Wow" Moment:

"Imagine your AI doing design research before generating UI. Studying real screens and user flows instead of guessing" — @bbssppllvv, 1,912 likes, 146K views

Real User Feedback:

"Love the filter functionality the most — very helpful and easy to navigate and be inspired!" — ProductHunt User "would not have to spend hours looking at 100 different websites to get design ideas" — ProductHunt User

For Indie Developers

Tech Stack

• Protocol Layer: MCP (Model Context Protocol) — Created by Anthropic, donated to the Linux Foundation in Dec 2025
• Interface: MCP Server, exposing 5 tools to the AI Agent
• Data Layer: 124,000+ real product screenshots, 45+ page types, 36 user flow categories, 87 UX patterns
• Compatible Clients: Cursor, Claude Code, Goose, and any Agent supporting MCP

Core Implementation

Refero MCP's technical logic boils down to three steps:

1. AI Agent receives an instruction like "design a dashboard"
2. Agent calls Refero tools via MCP (search_screens, search_flows, get_design_guidance) to see how real-world dashboards are built
3. Agent generates UI based on real references rather than hallucinating from scratch

This is the "Research-First design methodology" — AI researches before designing, just like a real designer studies benchmarks before drawing.

Open Source Status

• Closed-source commercial product, no public GitHub repository
• Skill package registered on agentskill.sh (@neversight/refero-design)
• Build difficulty: Medium-High. The challenge isn't the tech, but the data — crawling, cleaning, labeling, and updating 120,000 screenshots. The tech part (MCP Server + Search API) could be done in 1-2 months, but the data moat took Refero 3 years to build.

Business Model

• Subscription SaaS: Pro $12-14/month, annual approx. $8/month (33% discount)
• MCP included in Pro subscription at no extra cost
• 3-day free trial, no credit card required
• Annual revenue €1.63M (2022 data, pre-MCP)
• 30% discount code provided during PH launch

Giant Risk

Medium risk. Figma already has an official MCP Server, but it connects to your own design files, not a reference library. Theoretically, Figma or Mobbin could do this, but they haven't yet. Refero's moat is its 120,000+ continuously updated screenshot database + first-mover advantage.

For Product Managers

Pain Point Analysis

• Problem solved: AI-generated UI is repetitive and looks like it all came from the same template
• Severity: High-frequency demand. With the rise of Vibe Coding, more people are using AI for frontend, but quality is the bottleneck. As Mike Bespalov says: "Most AI-generated interfaces look the same. Not because the models are bad. Because they've never studied real product design."

User Personas

Feature Breakdown

Competitor Comparison

Key Takeaways

1. "Research-First" Philosophy: Let AI research before executing. This can extend to any AI tool — e.g., AI studying competitor copy before writing, or AI checking similar open-source projects before coding.
2. MCP as a Product Extension: Exposing an existing database to AI via MCP gains new users at almost zero marginal cost. This is the "API-first" strategy of the MCP era.
3. Small Team + Data Moat: 3 people, €1.63M ARR, built on a 3-year accumulation of 120,000 screenshots.

For Tech Bloggers

Founder Story

• Mike Bespalov (aka Mikhail Bespalov), background in product design
• Previously created Spektr (iOS photo editor)
• Designed medical apps at Sigma Software Design
• Former Clerk team member
• Founded Refero in 2022, starting with manual curation of design references for a year
• Based in Barcelona, describes himself as "just trying to build good things"
• 3-person team: Mike + Lidiya + Stepan Smirnov (Engineer) + Denis Demidov
• Bootstrapped to €1.63M annual revenue with zero funding

Controversies / Discussion Angles

• "Can AI actually have design taste?" — A great debate topic. Refero's answer is "feed it real references," but some argue it's just a different kind of templating.
• Data Copyright — Where is the legal boundary for 120,000 screenshots of real products?
• Will MCP make designers obsolete? — When AI can automate research + generation, what is the value of a junior designer?

Hype Data

• PH Ranking: 123 votes
• Twitter Buzz: Founder's post got 1,912 likes + 146K views (single post), very high reach for the MCP tool category
• Search Trends: "MCP design" is on an overall upward trend

Content Suggestions

• Angles: "The Design Quality Crisis in the Vibe Coding Era" or "Giving AI a Designer's Eyes with MCP"
• Trend Jacking: Vibe Coding, the MCP ecosystem, and AI toolchains are all hot topics for 2026

For Early Adopters

Pricing Analysis

Hidden Costs: Requires an existing MCP-supported Agent (e.g., Cursor Pro $20/mo, Claude Code pay-as-you-go).

Getting Started

• Setup Time: 5-10 minutes
• Learning Curve: Extremely low
• Steps:
  1. Sign up for Refero Pro (3-day free trial)
  2. Visit refero.design/mcp to get your MCP config
  3. Add Refero in Cursor/Claude Code MCP settings
  4. Use as normal — AI will automatically search references before generating UI
  5. Try demo.refero.design to see the effect first

Pitfalls and Complaints

1. No full user flows — Refero focuses on single screens. For full onboarding flows, Mobbin is superior.
2. Mixed search results — Screenshots from different products are jumbled together; lack of product grouping can be confusing.
3. Infinite scroll issues — The list gets too long; needs pagination or a "back to top" button.
4. Update lag — Refero might not immediately sync when a referenced product updates its design.

Security and Privacy

• Data Storage: Refero stores the screenshot database; your Agent queries via MCP. It does not upload your code.
• Privacy Policy: The MCP connection is for searching and reading references only; it doesn't touch your project data.
• Security Audit: No public audit report available.

Alternatives

For Investors

Market Analysis

• AI Design Tool Market: $4.4B (2023) → $26.5B (2033), CAGR 19.6%
• Generative AI Design: $994M (2025) → $16.9B (2035), CAGR 32.75%
• Agentic AI Market: $7.8B → $52B (2030)
• MCP Ecosystem: Monthly SDK downloads over 97M (Python + TS)
• Drivers: Vibe Coding explosion, rising enterprise AI adoption (Gartner: 40% of enterprise apps will include AI Agents by end of 2026)

Competitive Landscape

Team Background

• Mike Bespalov (Founder): Product designer, ex-Clerk/Sigma Software Design
• Core Team: 3-4 people, based in Barcelona
• Track Record: Spektr (iOS App), grew Refero to 250k users + €1.63M ARR within 3 years

Funding Status

• Funding: $0 — Completely bootstrapped
• Revenue: €1.63M (2022, pre-MCP)
• Valuation: Not disclosed
• Ranked 87th out of 991 competitors on Tracxn

Conclusion

The Verdict: Refero MCP hits the "generic AI UI" pain point perfectly. By using a "Research-First" methodology and the MCP standard, it lets AI study real products before designing. Small team, bootstrapped, clear data moat, and sharp positioning.

Resource Links

2026-03-16 | Trend-Tracker v7.3