AI builders default to generic primitives. Give the model a design language it can apply in one shot — UIcockpit exports the tokens and the prompt that wields them.
The reason AI-built apps look alike is that the model has no opinion to apply. UIcockpit fixes that: configure a system, then export a behaviour-shaping prompt that tells any AI tool which token, which variant and what sits next to what.
Drop the prompt into your tool’s rules file (.cursorrules, CLAUDE.md, a system prompt) and every generation respects the tokens, breakpoints, z-index stack and chart palette — automatically.
Not just variables — a rules file that teaches the model how to use them consistently.
Every kit ships a machine-readable contract and a verifier — uicockpit check, over a CLI and an MCP server. Wire it as a pre-commit or CI gate (--strict) and any drift — a raw hex, an off-scale radius, the wrong token — fails the build, so what your AI generates stays on-system across every screen and session.
Framework-neutral output and a portable prompt fit any AI coding workflow — plus an MCP server so agents install the kit, read its design context and check conformance natively.
Tokens (CSS + JSON), a Tailwind @theme block, a shadcn globals.css, an AI prompt and a machine-readable contract — all derive from one config, so they can never disagree. Plus a hosted live <link> you return to, and a CLI + MCP so your agent reads the system and checks its own work against it.
100% free, framework-neutral tokens — no account, no lock-in, your design language, your file.