The color engine, over MCP
Perfect Corp ships YouCam itself over MCP, so an agent can already ask it for the hex values of a face. What it cannot do is turn those into a season, a palette, or real garments ranked by measured color distance. That is what this exposes, and it needs no YouCam credentials of its own, because the engine is pure. It composes with their server rather than duplicating it.
Add it to a client
Claude Code: claude mcp add --transport http drape https://drape-youcam.vercel.app/api/mcp
Claude Desktop, in claude_desktop_config.json:
{
"mcpServers": {
"drape": {
"type": "http",
"url": "https://drape-youcam.vercel.app/api/mcp"
}
}
}Try the tools
These run against the same live server an agent talks to, over the same JSON-RPC. The skin color below, #be9c82, is a real reading YouCam returned for one of the sample faces, so the answers here match what the studio shows for it.
analyze_seasonTurns measured facial colors into a season, the three axes underneath it, the palette, and a confidence score that says how much to trust the read.
{
"skin_hex": "#be9c82",
"eye_hex": "#4f4030"
}find_garmentsRanks real garments by CIEDE2000 against the palette. Colors are measured from the retailer’s own fabric swatches, and sold-out colorways are left out.
{
"skin_hex": "#be9c82",
"gender": "female",
"limit": 4
}check_colorAnswers "does this specific shade suit me?" for a garment the shopper is already looking at. This one asks about a deep teal on a warm, light reader.
{
"skin_hex": "#be9c82",
"garment_hex": "#1f5f5b"
}correct_readRe-derives the season with an axis set by hand, for when the shopper disagrees. Deterministic, so it costs nothing and needs no new photo.
{
"skin_hex": "#be9c82",
"depth": "deep"
}Reading this as an agent rather than a person? The same thing in one markdown file: /llms.txt.
353 garment colorways back find_garments, each measured from the retailer’s own swatch image and re-checked daily against what they are still selling.