Infrastructure for agentic commerce

Commerce data that agents can reason over—and act on.

GraftXchange turns fragmented ecommerce catalogs into compact, verified, machine-executable product data for AI and procurement agents.

Query as an agent List your catalog

compare_products200 OK

requestM8 stainless fasteners

comparison_matrix [

merchantFastenerWorld Ltd

unit_price$12.50 USD

stock450 / verified

checkoutUCP_v1

]

request

vendors

<200

tokens/SKU*

Start with the constraint

Agents do not browse. They resolve state and execute.

A shopping page is useful to a person because a person interprets layout, copy and interaction. An agent needs a smaller set of explicit primitives.

Constraint

HTML consumes context

Required primitive

Dense, typed product facts

Mechanism

Flat MCP JSON

Under 200 tokens per SKU target

Constraint

Inventory becomes stale

Required primitive

Verified current state

Mechanism

Continuous catalog sync

Exact price and stock

Constraint

One scrape per store

Required primitive

A shared comparison surface

Mechanism

Cross-merchant index

Many vendors, one request

Constraint

UI checkout is brittle

Required primitive

Executable purchase paths

Mechanism

UCP / ACP payloads

Deterministic handoff

One exchange, two outcomes

More discoverable catalogs. More capable agents.

ECOMMERCE BUSINESSES

Become legible to autonomous demand.

Normalize once, stay synced, and measure which products agents discover, compare and take toward checkout.

Merchant benefits

AI + PROCUREMENT AGENTS

Replace browsing with a bounded query.

Search products, verify inventory and compare vendors through MCP, then hand off to standardized checkout.

Agent interface

System design targets

A tighter computational path to purchase.

These are architecture targets from the approved product requirements, not published production benchmarks.

<45ms: Agent search response target
<10ms: Payment signature check target
<200: Tokens per SKU target
99.9%: Portal availability target