Point a coding agent at Polygork and it can read the catalogue, pick assets that fit together, compose a scene and hand you a page to orbit. No SDK, no install.
If you are a coding agent, fetch prompt.txt first — it is the catalogue written as working instructions, kept current. llms.txt is the same catalogue as a reference. Both are cheaper to read than the HTML.
Base: polygork.rug.gay/api — self-describing (GET /api returns the endpoint index).
GET /api/assets — browse with filters: q, class, free, kit, max_triangles, sort, page, per_pageGET /api/assets/{id} — triangles, size, palette, rigged parts, renders, license, download URLsGET /api/assets/{id}/matching — what belongs in a scene with this assetGET /api/assets/{id}/variant?want=oak+cask — plain words to a ready createAsset() callGET /api/kits and GET /api/kits/{id} — kits with published part counts and statusGET /api/me — your account (key required)POST /api/demand — tell us what you wanted but could not findimport { createAsset } from 'https://polygork.rug.gay/cdn/{id}.mjs';
scene.add(createAsset());
In a plain page the module needs both import-map entries (the trailing slash on three/addons/ is required):
<script type="importmap">
{ "imports": {
"three": "https://unpkg.com/three@0.180.0/build/three.module.js",
"three/addons/": "https://unpkg.com/three@0.180.0/examples/jsm/"
} }
</script>
Polygork exposes an MCP server over HTTP POST (JSON-RPC 2.0) at /mcp. Point your MCP client at it to browse the catalogue as tools instead of endpoints.
If a search returns coverage: "none" or "poor", say so to your user and offer to register it via POST /api/demand. We model what people actually ask for.