Get started
Build with scentrev-mcp
14 MCP tools that turn community fragrance knowledge into clean, structured data. Ask in plain English - your agent handles the rest.
You ask · your agent calls the API
“Best beast-mode winter scent from Lattafa?”
No SQL, no slugs to memorize. Ask in plain English. Each tool below includes copy-ready examples for that exact use case.
Category cheat sheet
Exact filter labels for search_fragrances_filtered and blend tools: longevity, sillage, rating, appreciation, discovery sort, and more.
Generate a key
Sign in with Google, GitHub, or email, then create a key on the API Keys page. One active key per account.
Add the config
Paste one block into Cursor or Claude Desktop, or call the REST RPCs directly with your bearer token.
Ask away
Type a question in plain English. Your agent picks the right tool and returns structured longevity, sillage, notes, and sentiment data.
What you can ask
Three common flows. Pick a tool in the sidebar for more examples tied to that endpoint.
Discover
search_fragrances → search_fragrances_filtered
- “Find Creed Aventus.”
- “Best summer office scents with moderate sillage.”
- “Hidden gems from Lattafa with strong longevity.”
Analyze one scent
get_fragrance_profile · get_performance · get_appreciation
- “Full profile for Creed Aventus.”
- “Is Layton loud or intimate?”
- “Do people love Club de Nuit Intense?”
Compare & blend
get_reminds_of · search_by_references
- “What smells like Aventus but cheaper?”
- “Blend Aventus with Oud Wood.”
- “Alternatives to Lost Cherry.”
Connect a client
Point any MCP client at https://api.scentrev.com/mcp/ and authenticate with your bearer key. The same key works for direct REST calls at https://api.scentrev.com.
{
"mcpServers": {
"scentrev-mcp": {
"url": "https://api.scentrev.com/mcp/",
"headers": {
"Authorization": "Bearer frag_live_YOUR_KEY"
}
}
}
}Generate the key from the dashboard after signing in. Creating a new key revokes the previous one.
Install on Smithery
Connect through Smithery with your API key. Same upstream endpoint and tools as a direct MCP client config.
Output rules
Four invariants every response follows, so you can trust the shape of the data you build on.
Ratios only
Raw likes, dislikes, vote JSON, and 0-5 star averages are never exposed. You get normalized scores instead. Star rating is out of 5 - convert with score × 5, never × 10.
n_records, not confidence
Every metric carries n_records - how many community votes shaped it. Over 500 is a strong signal; under 100 is thin data.
0-1 everywhere
All wear metrics are normalized to 0-1 with a category label. The one exception: accord percentages run 0-100.
Rating vs reviews
Star rating uses n_records (rating votes) on a 0-5 scale. reviews_count is how many written text reviews exist - a separate number, not a /5 or /10 score.
A typical metric object
Score, category, vote count, and how to read the axis - every time.
{
"score": 0.62,
"category": "moderate",
"n_records": 1240,
"scale": "0 = X, 1 = Y"
}