What it is
getdesign.md is a catalog, not the origin of the DESIGN.md format. It presents 550+ reverse-engineered files from known websites, as a frontend for VoltAgent’s awesome-design-md collection. Its own about page says the files are reference material, not exact reproductions or official analyses by the listed companies.
The canonical format is Google Labs’ design.md repository. Google open-sourced it on 2026-04-21 as a format for describing visual identity to coding agents. The specification uses YAML frontmatter and a Markdown body. Its tooling ships lint, diff, Tailwind export, World Wide Web Consortium (W3C) Design Token Format export, contrast checks, and broken-reference detection.
First principles
A catalog is only as useful as the trust a reader can place in each entry. Scale is easy to see. Trust is not, and more entries do not raise it.
A reverse-engineered DESIGN.md is a reconstruction. A person or a tool examined a public site and recorded what it shows and what the rules appear to be. The file format splits that work in two: the frontmatter holds values a program can check, and the Markdown body holds rationale only a reader can judge.
The linter covers the first half. It confirms syntax, references, and contrast. It cannot confirm that a file matches the company it is named after.
So a catalog like this needs two things most: a visible label on each entry saying how it was made, and enough of each file on the listing page to judge it before copying it into a project.
Who is behind it
getdesign.md is maintained by the VoltAgent team. The project lists necati@voltagent.dev for sponsorship contact. As of September 2026, VoltAgent’s underlying awesome-design-md repository had 117,378 stars and was MIT licensed. VoltAgent’s main voltagent repository had 10,663 stars. Those counts show visibility, not that every reverse-engineered file is accurate.
The site names no individual founder. Attribution beyond the VoltAgent organization is UNVERIFIED.
Format and how an agent consumes it
The catalog listing shows thumbnails and labels; the content comes from the upstream collection and Google’s spec. A minimal canonical form is:
colors:
primary: "#1A1C1E"
typography:
h1:
fontFamily: Public Sans
fontSize: 3rem
components:
button-primary:
backgroundColor: "{colors.primary}"This is a hybrid artifact. The frontmatter carries structured values, and the Markdown explains how to use them. The common workflow is simple: copy a file into the project root beside package.json. Claude Code finds DESIGN.md automatically; Cursor needs an explicit @DESIGN.md reference. The upstream validator provides a standards-based check:
npx @google/design.md lint DESIGN.mdThe command reports rule results and Web Content Accessibility Guidelines (WCAG) AA contrast compliance. getdesign.md itself exposes no Model Context Protocol (MCP) server, browser extension, application programming interface (API), or command-line interface (CLI). Its agent experience depends on the plain-file convention and the upstream Google tooling.
Pricing and access
The public catalog is free to browse. The site also offers private or requested DESIGN.md work, a Catalog Pass subscription, premium starter kits, and an affiliate program. The Catalog Pass price did not render during research, because its widget loaded client-side. The exact price is UNVERIFIED. No dedicated pricing page was linked.
What works
The project clearly credits Google Stitch instead of claiming the format. That is a meaningful trust decision. It puts a large community catalog on top of a simple file convention, so users can evaluate an example and bring it to an agent without new infrastructure.
The upstream linter makes the file more than inspiration. A user can at least validate syntax, references, and contrast, instead of treating every Markdown file as equally sound. The separate official-design-md collection also shows that first-party files can be kept apart from community reconstructions.
What does not
The browsable site is thin at the list level. It shows thumbnails, not enough raw content to judge a file without opening it. More important, the files are explicitly not official reproductions. Their accuracy is a known limitation, not an incidental one.
The paid Catalog Pass price could not be verified. And getdesign.md adds no programmatic way to search or fetch files of its own. An agent must be handed a copied file, which is a workable baseline but not a fresh-data workflow.
What we take from it
Velofy could propose a catalog that clearly separates first-party, community, manually derived, and model-inferred entries. It could show enough of each spec to judge it before download, and link directly to the upstream standard and validator.
It could also propose a programmatic, versioned retrieval path. Velofy has not shipped an MCP server, CLI, or package. The point is to avoid presenting a proposal as if it were infrastructure already available.