Build recommendations

Perfume recommendation API for AI agents

A perfume recommendation API should filter on wear metrics — longevity, sillage, season, notes — not on marketing adjectives. Here is how to build AI perfume recommendations on scentrev-mcp.

Most “AI perfume recommenders” are a chat model plus a scraped list. They cannot enforce “lasts eight hours” or “office-safe sillage.” A real perfume recommendation API exposes filters the model must obey.

On scentrev-mcp that tool is search_fragrances_filtered (MCP or REST). You pass season, gender, min longevity category, min sillage category, notes, and a sort. The catalog returns at most ten rows with Bayesian rating scores — not a dump of the whole database.

Example recommendation intents

Summer office scent, moderate sillage, beloved appreciation.
Winter oud, eternal longevity, male or unisex.
Dupes of a luxury reference under a price ratio.
Blend two loved bottles and find the overlap in notes.

After you have candidates, call get_fragrance_profile or compare_fragrances so the agent can explain why it picked one. That explanation is what users trust; a naked name list is not a recommendation engine.

Wire it through the perfume MCP in an IDE, or through the fragrance API in your product. Same recommendation semantics either way.