Fudge MCP: AI agents can now search 10,000 real sites for design references

What happened

Fudge MCP launched on Product Hunt on July 13, 2026. It is a design reference engine that runs locally via the Model Context Protocol, giving AI coding agents access to a database of roughly 10,000 real production websites searchable by visual properties: font choices, color palettes, layout patterns, component structures, and visual similarity.

Context

The persistent problem with AI-generated UI is that prompts relying on adjectives — “modern,” “premium,” “clean” — produce output that is technically correct but generically styled. Fudge MCP addresses this by giving the AI agent measured design evidence with screenshots from real sites, rather than asking it to interpret subjective language. A team building a SaaS dashboard can point their agent at examples of dashboards that actually look the way they want, and the agent can reference specific visual patterns from those examples.

The tool comes with a Chrome extension for saving design references locally, and runs without any cloud dependency or account requirement. This means the reference database stays on the developer’s machine and is not sent to a third-party server.

Why it matters for designers

As AI coding agents take on more UI production work, the interface between designer intent and agent output becomes a practical constraint. Fudge MCP shifts that interface from natural language description toward concrete example. Designers who maintain a library of visual references can now make those references directly actionable in AI coding workflows, rather than translating them into words and hoping the agent interprets them correctly.

This is a small but meaningful shift in how design decisions travel through an AI-assisted production pipeline.