The Gap
Why This Matters
Anthropic's Claude Commerce Agents blueprint (released September 2, 2026) gives merchant agents the harness to draft pricing, promotion, and campaign changes for human approval — reading a retailer's own sales performance, stock, and order history. What it doesn't provide is visibility into the market outside that store. A merchant agent can recommend a price change with complete confidence in its own numbers and still be wrong the moment that price is compared to the market. kwery closes that gap as a callable tool: GTIN-matched pricing across 12 marketplaces, with a confidence score on every result, returned in one request — no polling, no batch job, no separate pipeline to maintain.
Setup
Get a Key and Test Against the Mock Server
Sign up at kwery.co — your organization is created with a free credit allowance, and the dashboard issues an API key (kwy_live_…) that authenticates every request. Send it as a Bearer token or the x-api-key header.
Before touching production, prototype the tool definition against kwery's mock server — no signup, no key, no credits consumed. It returns real response shapes so you can build and test the agent's handling of the result end to end, then swap the base URL when you're ready to go live.
curl -X POST https://docs.kwery.co/_mock/openapi/job \
-H "Authorization: Bearer test123" \
-H "Content-Type: application/json" \
-d '{"source":"amazon","country":"de","key":"gtin","values":["4006381333962"]}'Implementation
Define the Tool Call
Map the tool's parameters to what the merchant agent's harness already has available — a product identifier from the retailer's catalog, and the marketplace to check it against:
Production call:
curl -s https://api.kwery.co/job \
-H "Authorization: Bearer $KWERY_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"source": "amazon",
"country": "de",
"key": "gtin",
"values": ["4006381333962"],
"wait": true
}'This is the same request shape as any other kwery integration — nothing agent-specific about the API itself, only about how you wire the result back into the harness's decision loop.
Response Contract
Handling the Response
Always check the error field first — validation problems (bad identifier, unsupported source/country combination) return HTTP 200 with "error": true, not a failed status code. Only infrastructure problems (auth, server errors) use standard error status codes.
A successful result carries key, success, and the parsed content for that value — including the matched price, offer data, and the confidence-score breakdown (identity, completeness, price_integrity, freshness). Feed the price and its confidence score back into the agent's pricing-recommendation step together — a low-confidence match is a signal to flag for human review rather than act on automatically, which fits naturally into the blueprint's existing human-approval gate for pricing changes.
Going Further
Scaling Beyond Single Lookups
The wait pattern above is built for one-off checks — an agent pricing a single SKU mid-conversation. For a merchant agent doing a catalog-wide sweep instead of a single lookup:
Batch API
Submit up to 1,000 identifiers, poll, then download results. Good for a scheduled or on-demand full-catalog price check.
Stream API
Submit up to 100,000 identifiers, results pushed to your endpoint via signed webhooks as they complete. Good for large catalogs where you don't want to hold a connection open.
Full details in API Documentation.