What it is
TypeUI is an open-source command-line interface (CLI), hosted Model Context Protocol (MCP) server, and public registry for design skills. Its unit of distribution is a SKILL.md file, often paired with DESIGN.md. The registry holds 70+ named skins such as Minimal, Brutalism, Glassmorphism, Neon, Claymorphism, Dithered, and Editorial, plus a fundamentals skill and 17 categories of user interface (UI) prompts. These are style presets, not tokens extracted from a particular product.
The distinction matters. TypeUI is useful when a team needs a coherent starting direction. It is not evidence of how a known product is built. The hosted site was behind a Vercel security checkpoint during this research, so site-level facts that came only from search snippets are treated cautiously here.
First principles
A design skill works only through the text it puts in front of an agent. Two things decide its effect.
First, the agent has to load it. That is a distribution problem: the file must sit where the agent looks, or be reachable through a tool the agent can call. Second, the text has to be specific enough to override the model’s defaults. A named font and a spacing scale give the model something exact to follow. An adjective like “minimal” leaves most choices to the model.
TypeUI’s CLI, MCP server, and registry address the first problem. Its populated skins address the second with concrete values, as the example below shows.
What a preset cannot carry is evidence. A skin is written to a style, not measured from a product. It can make output consistent. It cannot make output match an existing brand, because nothing in it came from that brand.
Who is behind it
The project is from Bergside LLC. The repository’s license and plugin metadata name Zoltán Szőgyényi as founder and maintainer. Skybridge is named as a README sponsor. The project is MIT licensed. As of September 2026 its GitHub repository had 1,967 stars and 139 forks. Those are repository facts, not a measure of adoption quality.
The Pro tier is described as enhanced skill files, more layout prompts, a private Discord community, and monthly updates. Its exact price was not available because the pricing page could not be fetched. That price is UNVERIFIED.
Format and how an agent consumes it
The core is Markdown with YAML frontmatter. This is the project’s published DESIGN.md blueprint:
---
name: design-system-[brand-or-scope]
description: Creates implementation-ready design-system guidance with tokens, component behavior, and accessibility standards. Use when creating or updating UI rules, component specifications, or design-system documentation.
---
<!-- TYPEUI_SH_MANAGED_START -->
# [Design System Name]
## Mission
One paragraph describing the system objective and target product experience.The rest of the template covers brand context, visual style, typography, colors, spacing, radius, shadows, motion, and accessibility. A populated skin can be much more concrete:
## Style Foundations
- Visual style: minimal, clean
- Typography scale: 14/16/18/24/32/40 | Fonts: primary=Roboto, display=Montserrat, mono=PT Mono
- Color palette: primary, neutral, success, warning, danger | Tokens: primary=#111111, secondary=#8B5CF6
- Spacing scale: 4/8/12/16/24/32An agent can get this material in several ways. The typeui.sh npm CLI can generate, update, randomize, list, or pull skills. The hosted MCP server at https://mcp.typeui.sh exposes the registry to agents. A generic skill install also works:
npx skills add https://github.com/bergside/typeui --skill typeui-fundamentalsThere are installation guides for more than 20 agents and integrated development environments (IDEs). TypeUI also ships a Visual Studio Code (VS Code) and GitHub Copilot marketplace plugin. The registry index is deliberately simple JavaScript Object Notation (JSON), mapping a slug to skillPath and designPath.
Pricing and access
The CLI and public registry are free and MIT licensed. Pro is a paid subscription through Polar.sh, according to the research. The reported difference is a single-file free skin versus 20+ Markdown files per Pro design skill, plus more prompts, community access, and updates. The exact subscription price is UNVERIFIED because the source page was unavailable.
What works
TypeUI has a practical distribution shape: raw Markdown for agents, a CLI for local work, and MCP for live lookup. Separating SKILL.md rules from DESIGN.md rationale is also sound. An agent gets a compact behavioral instruction, while a person can read the longer explanation. The flat registry is easy to inspect and needs no proprietary package manager.
The project also shows that one source of content can serve a broad set of tools. A team can pick its editor or agent without learning a different design vocabulary for each.
What does not
The skins are generic aesthetic directions. “Neon” and “Cafe” do not answer the harder problem of faithfully matching an existing brand. The quality and consistency of 70+ hand-written skins are UNVERIFIED. No evaluation method or before-and-after evidence was found.
There is also a practical contradiction. A tool built around agent-readable design guidance had a marketing site that blocked automated retrieval during research, and its Pro price is opaque. Neither is fatal, but both weaken the pitch.
What we take from it
Velofy could propose a dual artifact: an agent rule file and a human explanation that point at the same design system. It could also propose one registry feeding a CLI, an MCP endpoint, and explicit setup instructions per tool. That is a distribution pattern, not a shipped Velofy package or MCP server.
The stronger proposal is clear provenance. TypeUI’s presets are useful as presets, but they should not be confused with values taken from a real product. A future Velofy catalog could label that distinction directly, and publish its price and access terms where agents and people can actually read them.