Klevu (E-commerce AI Search) MCP. Control product discovery—from keywords to trends.

You're connecting Klevu to your agent. This server lets you ditch basic site search and actually take control of product discovery across your whole catalog. You gotta run sophisticated searches that feel natural, whether you're auditing merchandising paths or just trying to find a specific widget.

When you start talking about products, the system doesn't guess; it pulls data directly from ten distinct methods. You can use search_keyword to run a general product search across your entire catalog using natural language keywords, giving you high-relevancy results right out of the gate.

Before you even type anything, you get help with search_autocomplete, which grabs real-time suggestions for search terms as a user types into the bar. If you need to check how items are displayed in a specific section, you use search_category to retrieve all products assigned to that defined category page path, letting you audit merchandising rules exactly where they live.

When your general keyword search gets too broad, you refine it immediately using search_filtered. This tool runs a focused search by applying explicit attributes—you can narrow the results down instantly by specific filters like color, size, or brand. You'll also use search_sorted, which runs a keyword search but lets you dictate the exact order the results should appear in; for example, you can make sure everything is sorted by price.

Since catalogs are huge, you gotta handle volume with search_pagination. This tool retrieves result sets for any query and allows your agent to cycle through multiple pages of products so you never hit a dead end. If all that fails, or if the search structure gets really weird, you can bypass everything and use search_raw.

This executes any custom search payload you build using raw JSON directly against the Klevu API for deeply customized query testing.

For pinpoint accuracy, you've got search_product_id. If you know a product’s unique catalog ID number, this tool gets all the specific details for that single item. You don't have to guess; you just check the ID and get the full record.

On top of finding what exists, your agent can predict what people want next. search_recs fetches suggested products based on behavioral data and machine learning models—it shows what shoppers might like or frequently buy together. You also use search_trending, which views exactly which products are currently spiking in popularity and relevance across the whole store, giving you insight into seasonal spikes.

You're not limited to finding things by text or category structure; your agent can build a complex query using search_filtered for color, size, and brand simultaneously. You pull product lists from defined paths using search_category, and then you narrow those results down further with attributes via search_filtered. The entire process is designed to take vague user intent—like 'nice shoes'—and turn it into a precise list of inventory, page by page, sorted exactly how you want it.

You can get the full product record using search_product_id regardless of whether it was found through a keyword search via search_keyword, or if your agent is following up on recommendations pulled from search_recs. The system handles both general searches and highly structured ones, ensuring you always see the most current data available.