Cre8 app factory
Two plugins · packaged for Claude, ChatGPT, and Gemini

The app factory that makes your company AI-native.

Cre8 is a design system built to be read by machines — and two plugins that turn it into a factory floor. Inside Claude, ChatGPT, or Gemini, your agents open real backend connections to your databases and services, compose what comes back into 93 accessible on-brand components, and verify the result actually renders. The output is applications your company can ship, not prototypes of them.

93components, WCAG-accessible
2plugins — interface + backend
15MCP tools
frameworks — it’s the platform
generative session
U
prompt“Build me a getting-started card with a heading, a couple of highlight badges, and a primary CTA.”
C
Cre8-aware agentQuerying the design system…
search_components("card, badge, button")
get_content_model("cre8-card")
Composing verified components.
Rendered output
Getting started

Everything you need to ship your first generated surface.

AccessibleThemeable
Primary action

See these rendered as live components on the components page.

Two plugins

Packaged for Claude, ChatGPT, or Gemini.

An application is a backend and a screen. Generated UI that isn’t wired to anything is a mockup, and a query nobody can act on is a dead end. Installed together, these two give one agent the entire job — opening real, credentialed connections to your databases and services, then rendering what comes back as accessible, on-brand, production components.

ClaudeClaude Desktop and Claude Code
ChatGPTChatGPT Desktop and Codex
GeminiGemini and Gemini CLI
The interface half

cre8

Grounds the agent in a real, governed component library, so what it emits is production markup rather than a plausible-looking approximation of it.

  • 93 WCAG-accessible components, served by 15 MCP tools
  • 5 skills covering composition, token architecture and framework conventions
  • 3 specialist agents and a verification workflow that refuses unproven passes
Explore cre8
The backend half

cre8-data

Stands up the actual backend connection — writing the SQL, reaching your warehouse and services under real credentials, and validating what comes back before anyone builds a decision on it.

  • 9 data and database skills, from SQL and schema profiling through validation and dashboards
  • 4 connected sources configured on install, plus typed contracts generated from an OpenAPI spec
  • Clerk machine authentication, so a gated system is a real connection rather than a mock
Explore cre8-data
Why the output ships

A factory is only as good as the products it assembles.

Most design systems are built for humans to read the docs. Cre8 is built so a model can, too — every component ships the metadata, guardrails, and structure an agent needs to get it right on the first pass. That is the difference between generated screens you ship and generated screens you demo.

Machine-readable by design

A full custom-elements.json manifest describes every tag, prop, slot, and event — so an agent knows exactly what exists and how to use it. No scraping, no guessing.

Composable component trees

Agents emit structured component trees, not brittle hand-rolled HTML. The output is real Cre8 elements — inspectable, diffable, and identical to what a human would ship.

Accessible by construction

Whatever the model generates inherits WCAG 2.1 semantics, keyboard support, and ARIA from the components themselves. Accessibility isn't prompted — it's guaranteed.

On-brand through tokens

Generated UI reads from design tokens, so a swap of the token file re-themes everything the AI produced. Cre8, brand, or a custom theme — the model stays inside the lines.

MCP-native

The @tmorrow/cre8-mcp server plugs component intelligence straight into Claude, Cursor, and any MCP client — live, versioned, and always in sync with the library.

Framework-agnostic

Web Components run anywhere — React, Vue, Angular, or plain HTML. Generative UI you build with Cre8 isn't locked to one renderer or one runtime.

The line

From prompt to production in four steps

The same pipeline behind a generative-UI playground — grounded in a real, governed component library, and wired to your own systems rather than sample data.

01 · ask

Describe the UI

A user or agent states intent in plain language — “a dashboard row with three stat cards.”

02 · ground

Query the system

The model calls the MCP server to find the right components, props, and approved patterns.

03 · compose

Generate real components

It emits a tree of verified Cre8 tags — typed, accessible, and token-themed by default.

04 · render

Ship it live

The browser renders native web components. What the AI drafted is what production runs.

Make your company
AI-native.

Install both plugins and your agents start building complete, accessible, on-brand applications against your own data — in the tools your team already has open.

Claude Code
// Both plugins, one marketplace
claude plugin marketplace add tmorrowdev/tmorrow_ai
claude plugin install cre8@tmorrow_ai
claude plugin install cre8-data@tmorrow_ai