Not as a packaged connector — and that's by design, not a gap. kwery is a plain REST/JSON API, so it already works from Claude, ChatGPT, LangChain, or any agent framework that can call an HTTP endpoint. A dedicated MCP connector would lock the integration into one specific path; staying a plain API keeps it usable everywhere without a separate wrapper to maintain.
kwery isn't an official Anthropic partner on this, but it fits naturally: the blueprint's shopping agent expects a product-search/catalog tool behind it, and kwery's API is a straightforward fit to wire in as that tool's data source for cross-marketplace price and offer lookups — you connect it yourself using the blueprint's existing tool-contract pattern. Read the integration guide.
Yes — a request returns the resolved price synchronously in the same response; there's no job-status endpoint to poll for a single lookup. Webhooks exist separately for streaming/monitoring use cases, not because single lookups need them.
A feed hands you whatever was last scraped on someone else's schedule. kwery resolves live on each request — cached results are only served within a freshness window you set yourself (up to 3 days), so a call outside that window always triggers a real fetch, not stale data with no way to know how old it is.
Structured JSON with a consistent schema across every source — no HTML to parse, no per-marketplace response format to normalize yourself.
Depends on the source, and most support more than one: GTIN and free-text search work across most marketplaces, plus source-specific identifiers where they exist (ASIN for Amazon, product ID for others).
Yes — each offer includes availability alongside price and merchant, wherever the source itself exposes it.
Every result carries a confidence score, not just a number — a measure of how reliable that specific result is, so an agent making a real purchasing decision has something to reason about beyond "here's a price." How confidence scoring works.
Yes — it's a standard REST/JSON API underneath, so it's callable from any framework that can make an HTTP request: LangChain tools, OpenAI function calling, or a custom agent loop, not just Claude.
A clean, structured response, not an error — "not found" is a normal outcome your agent's code can handle the same way as any other result, no special exception handling required.
Yes — kwery ships a mock server that returns real response shapes with no API key and no credits consumed. Prototype a tool definition, test how your agent handles the response, then swap in a live key when you're ready for production. API documentation.
The same credit model as everything else — one credit per successful or fresh "not found" result, regardless of whether the call came from a script, a dashboard job, or an autonomous agent. There's no separate agent pricing to reason about. View pricing.
Request API access and the team will walk through sources, volumes, and integration in detail.