Fudge MCP

Fudge is an MCP (Model Context Protocol) server that connects AI coding tools to a curated library of real website design references, helping AI generate websites that look designed rather than generated. Built by Withfudge, it is available as a subscription service with a companion Chrome extension for capturing new websites on the fly.

What It Is

Fudge sits between your AI coding assistant and a library of over 1,675 real websites, giving the AI concrete design evidence—screenshots, typography, color tokens, spacing, borders, radii, shadows, and layout relationships—before it writes UI code. Instead of relying on generic design advice, the AI can search the library by style, page type, font, color, or screenshot similarity, then carry those references directly into the coding conversation. The tool is positioned as a design intelligence layer for AI-assisted development workflows.

How the Library Works

The Fudge library, according to the product page, contains 10,094+ captured sections from real websites, 1,476+ font families, and contributions from 2,552+ designers. Each capture stores:

• The page: screenshot, source URL, page role, sections, components, layout, viewport, and page state
• The design system: typography, color roles, contrast, spacing, borders, radii, shadows, gradients, and component styles
• The experience: images, video, crops, playback, motion, interactions, overlays, and clipping

Unlike editorial galleries such as Mobbin, Refero, or Dribbble, Fudge lets users add any live website immediately via the Chrome extension without waiting for a curation queue.

Core AI Skills Added

Once installed, Fudge adds a set of design-specific tools to any compatible AI client:

• Font search – find typefaces by name and see production pages using them
• Font similarity – find visually similar font families with real examples
• Visual similarity – find pages that match a reference screenshot's feel, density, or contrast
• Color operations – lighten, darken, mix, convert to OKLCH, and run WCAG or APCA contrast checks
• Categorized designs – search by UI pattern (pricing pages, dashboards, login pages, docs, modals, etc.)
• Chrome capture – extract design evidence from any website captured with the extension
• Design system export – turn captured evidence into Tailwind v4, CSS variables, DTCG tokens, or raw JSON
• design.md generation – summarize selected references into AI-readable design notes

Setup Path and Compatibility

Installation runs as a single command that auto-configures detected MCP clients. The product page lists automatic detection for Codex, Claude Desktop, Claude Code, Cursor, OpenCode, Windsurf, VS Code, Gemini CLI, and Zed. Manual JSON configuration is also available for other local-stdio MCP clients such as Warp, Kiro, and Roo Code. The Chrome extension handles website capture and feeds new references into the user's personal library.

Why It Matters for Vibe Coding Workflows

AI coding tools like Cursor, Windsurf, and Claude Code can generate functional code quickly but often produce visually generic results. Fudge addresses this by giving the AI a concrete visual starting point—real typography pairings, color palettes, spacing systems, and layout patterns from production websites—before any code is written. The export step (Tailwind, CSS variables, or JSON) closes the loop by turning design references into immediately usable theme files, reducing the gap between "I like this" and working code.