Trusted merchant knowledge for AI agents — verified, cited, monitored.
Shopping agents increasingly rely on merchant websites for product, pricing, and policy information — through model training, information retrieval, and real-time shopping data. That makes a brand's website a critical source of truth for AI, but the merchant information itself may be contradictory, outdated, or incomplete. When an agent gets a money fact wrong, it is often doing exactly what it was designed to do: faithfully repeating a real page — one that another page contradicts, or that changed after the agent's copy was made.
Contezza AI makes merchant knowledge trustworthy for AI agents. Contezza continuously evaluates merchant content, identifies contradictions and stale information, and maintains a verified knowledge graph built from the brand's own published pages: every answer served verbatim with citations, disagreements between a brand's own pages stated explicitly instead of silently resolved, and sources re-checked on a schedule so a stale answer is flagged before it can mislead a shopper.
Anthropic's commerce-agents blueprint agent, unmodified — the only change is the data source behind it. Asked about a brand whose FAQ and returns page state different restocking fees:
Raw pages"The returns policy states a 15% restocking fee…"
Verified feed"The store's published pages currently show two different figures — the FAQ cites 20% and the returns policy page cites 15% — and that discrepancy hasn't been resolved yet…"
Reproduced across independent runs. The first answer is confident and contradicted by the brand's own site; the second states the contradiction and doesn't guess.
From our brand audits (initial sample of DTC ecommerce brands, human-verified):
Contezza AI for Merchants exposes five read-only merchant-utility tools over governed, citation-carrying data:
check_policy_consistency — paste two policy texts (FAQ vs returns page) and
get a deterministic verdict on conflicting money facts — fees, return windows, warranty
and shipping terms — with the mismatched excerpts. Processes only the pasted text;
stores nothing.get_agent_readiness_checklist — concrete actions across the five dimensions
we audit (find, understand, repeat, visibility, monitoring), each grounded in audit
findings.get_brand_graph — a brand's knowledge graph: products, policies, and claims
with per-assertion citations to the brand's own published pages and verification
statuses. Where a brand's pages disagree, the contradiction is surfaced with both
sources instead of picking a side.get_category_norms — anonymized observed ranges and findings from audited
merchant cohorts, always with explicit sample framing.list_verified_brands — the brands whose knowledge graphs we serve:
audited pilot brands (every assertion cites the brand's own published pages),
demonstration data clearly labeled.Every answer is served from governed data — never generated — and every assertion carries a verification status (verified, needs reconciliation, or derived) with its source citation. The umbrella endpoint additionally serves Contezza's own verified answers and the audit-request pointer.
get_brand_graph and get_category_norms serve projections of
Contezza's Merchant Knowledge Graph — a governed graph of merchant facts, not a document
index:
Truth lives in a governed system of record; the connector serves versioned, read-only projections of it. Additional graph tools (typed traversal) are in beta.
Add the connector endpoint as a custom connector in your assistant's settings, or point any MCP-capable agent at it. No authentication is required for the public connector; brand tenants receive private endpoints with per-tenant keys.
Contezza AI (umbrella): https://mcp.contezza.ai/mcp Contezza AI for Merchants: https://mcp.contezza.ai/ecommerce/mcp Contezza AI for Financial Services: https://mcp.contezza.ai/finserv/mcp Transport: streamable HTTP · no authentication
One base host, many surfaces: the umbrella serves every public tool; domain surfaces (like Merchants and Financial Services) serve their domain's tools under their own identity. New domains appear as additional paths.
Added the Contezza AI for Merchants plugin from Anthropic's directory? Adding a plugin does
not connect its server. Open the plugin's Connectors tab, select
Add, then Connect, and switch it on in the chat under
+ › Connectors. On a Team or Enterprise plan an Owner adds it for the
organization first. In Cowork, if a session reports mcp.contezza.ai as blocked,
an administrator allows that domain in the organization's network settings.
Building on Anthropic's commerce-agents blueprint? Our open adapter serves the verified
knowledge feed through the blueprint's own backend contract — see the
contezza-for-commerce-agents
repository for a drop-in StorefrontBackend.
Connector tools are read-only. We log tool calls (tool name, query, timestamp) to operate and improve the service; details in our privacy policy.