Refero is a primary example of how niche, high-quality datasets are being exposed to the AI ecosystem via the Model Context Protocol (MCP). By providing a structured bridge to over 120,000 real-world product screens, Refero allows agents to move beyond generic UI generation. It occupies the knowledge or context layer of the agent stack, specifically targeting agents involved in frontend development and product design.
For developers, Refero’s "Skill" and MCP server serve as a design methodology for LLMs. This helps solve the problem of generic, "AI-looking" interfaces by grounding the model’s output in patterns from companies like Wise or Plaid. This push toward "designed, not generated" output is a key contribution to the agentic design workflow, making Refero an essential resource for those building autonomous UI/UX agents.
Refero sits at the intersection of design curation and the Model Context Protocol (MCP). While the platform began as a searchable repository of real-world UI and UX patterns for human designers, it has recently moved into providing a data layer for AI agents. Based in Barcelona and founded in 2022, the company maintains a library of over 128,000 screens from high-profile web and iOS products like Wise, Sana, and Monday.com.
The core utility of Refero for a human designer is research. Unlike many design galleries that feature concept designs which may not be functional, Refero captures live products. This allows users to see how major tech companies handle specific user flows — such as account deletion, onboarding, or complex dashboard navigation — in production. The library is categorized by page types, UX patterns, and UI elements, allowing for granular searches that bypass the need to manually sign up for dozens of competing services to see their internal screens.
The most significant development in Refero’s roadmap is its embrace of AI agents. Large language models are often trained on massive, unstructured datasets that include significant amounts of poor design. When an agent is asked to generate a user interface, it often defaults to generic patterns or layouts that do not follow modern usability standards. Refero addresses this by providing an MCP server. This allows an agent — such as Claude or a custom-built tool — to query Refero’s library of proven designs in real-time. By studying how established products handle specific components, the agent can generate code or layouts that reflect actual industry standards.
In the competitive landscape, Refero competes with established design galleries like Mobbin or Godly. However, Refero’s focus on structured data and "flows" — the step-by-step progression from one screen to the next — distinguishes it from static galleries. Its integration strategy is also more aggressive toward the professional workflow; while many competitors stay in the browser, Refero offers a Figma plugin and the MCP server to meet users where they work.
The company operates on a freemium model. A limited portion of the library is available for free, while the Pro and Team tiers provide full access to the Web and iOS collections, the Figma plugin, and the AI agent tools. This pricing structure suggests a target audience of professional product designers, growth teams performing competitor benchmarking, and developers building AI-driven design tools.
Refero remains a small team, with headcount estimated at under ten people. Despite its size, its entry into the MCP ecosystem positions it as a specialized knowledge provider. In the broader AI agent stack, Refero is the context layer — a curated dataset that grounds agentic behavior in real-world logic. As agents move from writing simple scripts to designing full application interfaces, the availability of these structured design references becomes a requirement for high-quality output.
A curated collection of real-world design references and user flows for web and iOS.
Refero is hiring