Merchant product catalog API — AI agent accessing digital inventory

UCP Product Catalog API: How to Expose Your Inventory to AI Agents

Last Updated: November 10, 2026 • Autonomous Agentic Commerce Spec v2026-08-25

Production Validated

The product catalog is the foundation of agent commerce. An agent cannot purchase what it cannot accurately understand. Most existing product data — optimized for human browsing through search terms, lifestyle photography, and curated descriptions — is poorly structured for agent consumption. UCP’s product catalog API defines a standard schema that makes merchant inventory legible to any AI agent, regardless of the underlying commerce platform.

Why Human-Optimized Product Data Fails Agents

Catalog Vector Human-Optimized Product Page Agent-Optimized UCP Catalog API Data Structure TTL & Cache Policy Agent Execution Velocity
Attribute Granularity Marketing prose, promotional banners, and visual lifestyle photos. Structured technical specifications (dimensions, materials, compatibility). Normalized Schema.org Product JSON-LD graph. ETag validation with max-age=300 seconds. Immediate programmatic suitability filtering.
Inventory Accuracy Vague ‘In Stock’ / ‘Limited Stock’ text badges. Atomic available-to-promise (ATP) inventory count by warehouse SKU. Strict integer inventory count with reservation delta. Real-time sub-10 second cache invalidation. Zero checkout failures due to phantom out-of-stock items.
Variant Resolution Complex multi-step dropdown selections requiring DOM scraping. Deterministic variant tree indexed by SKU and GTIN/UPC. Multi-dimensional variant matrix in single JSON payload. Cached at edge with ETag conditional validation. Sub-15ms multi-variant comparison across entire catalog.
Price Freshness Static HTML pricing subject to unexpected repricing at checkout. Dynamically signed quote endpoint with explicit expiration timestamp. Signed price payload with currency, discounts, and taxes. 15-minute cryptographically locked quote window. Eliminates price drift, cart abandonment, and customer dispute risk.
Catalog Querying Lexical keyword search returning paginated HTML results. High-dimensional vector embeddings + structured attribute filters. Vector embedding index paired with JSON-LD metadata. Edge-cached at CDN nodes worldwide. Sub-50ms intent-to-catalog matching for autonomous agents.

Human shoppers interpret ambiguous product descriptions, infer specifications from context, and use images to fill gaps in textual information. Agents do not. When a product description says large in the title and does not specify dimensions in structured fields, a human might look at the image and make an informed guess. An agent will either fail to match the item to the buyer’s requirements or purchase the wrong size. Structured, machine-readable product data is not a nice-to-have for agentic commerce — it is the minimum viable catalog.

UCP Catalog Schema: Core Fields

The UCP catalog schema requires specific fields that most commerce platforms support but rarely enforce: a canonical product identifier (not a variant SKU, but a root product ID), complete structured attributes (dimensions, weight, material, compatibility), a machine-readable condition field (new, refurbished, open box), availability status with restock date if applicable, and pricing structured as base price plus a complete fee schedule including shipping estimates by zone. Every field has a type and a validation rule. Ambiguous data is rejected at catalog ingestion, not at purchase time.

Variant Architecture for Agent Commerce

Product variants — size, color, configuration — require explicit parent-child architecture in the UCP catalog. Each variant must carry the full attribute set that distinguishes it from sibling variants. An agent selecting a size medium blue widget must be able to identify that specific variant unambiguously, add only that variant to the cart, and confirm that the variant is available at the stated price before completing the purchase.

Catalog Freshness and TTL

Stale product data causes agent errors. A price or availability status that was accurate when the agent cached the catalog and is no longer accurate when the purchase is attempted creates transaction failures that are frustrating for the human behind the agent and damaging to the merchant’s agent commerce performance metrics. UCP catalog records include a mandatory TTL (time to live) field specifying how long the record can be cached before requiring a fresh fetch.

Frequently Asked Questions

Why do traditional HTML ecommerce pages fail when crawled by shopping agents?

HTML product pages bury technical attributes in promotional text, lack deterministic variant identifiers (GTIN/UPC), and fail to expose atomic real-time inventory counts, leading to agent hallucinations and checkout errors.

How does the UCP Catalog API prevent race conditions on scarce inventory?

Agents request a temporary 5-minute inventory reservation lock during the cart phase, decrementing the available-to-promise (ATP) count atomically before initiating the payment settlement handshake.

What is the required data format for UCP product catalog endpoints?

All UCP catalog endpoints return RFC-compliant JSON-LD structured using Schema.org Product and Offer vocabularies, augmented with UCP-specific extension fields for machine intent matching.


Posted

in

by

Tags:

Comments

Leave a Reply

Your email address will not be published. Required fields are marked *