The worked examples in its own description set the granularity to expect — "iphone 15" as product research or purchase intent, "python tutorial" as learning intent — which is a category, not a ranked result page. Because the response arrives as named fields, it sits naturally in front of whatever actually retrieves, deciding which path a query takes. The label vocabulary is learned by calling it: the description fixes the response fields, and leaves the set of intents open.
A single-tool classifier for search queries: search_intent_analysis takes a `query` and returns its intent, related categories and suggestions.
- Intent classification with the vendor's own worked examples: "iphone 15" comes back as product research or purchase intent, "python tutorial" as learning intent.
- A named-field response rather than prose — query, intent, categories and suggestions — so the result can be branched on without parsing.
- Routing material alongside the label: the description lists topic categories, search suggestions and reference links among what it offers.
A Search Intent API key, supplied as SEARCH_INTENT_API_KEY.
One command plus a key — npx -y @search-intent/mcp, then supply credentials