How to Create Agent-Readable Product Feeds

Why Agent-Readable Product Feeds Matter in 2026

Agentic commerce surfaces — ChatGPT Shopping, Perplexity Shopping, OpenAI Operator, Apple Intelligence purchase flows, and emerging x402-based agent marketplaces — read product data through structured feeds and APIs, not by parsing your retail website. Brands without an agent-readable product feed are reachable to human shoppers but invisible to agentic buyers. This guide walks through publishing one.

Step 1: Audit Your Current Data Sources

• Google Merchant Center feed (if you sell on Google Shopping).
• Schema.org Product JSON-LD on PDPs.
• Product API (REST/GraphQL) if you have one.
• Catalogue export (CSV or JSON) used internally.

Most brands have 2-3 of these but no unified agent-readable feed. The goal is one canonical feed that agentic surfaces can consume reliably.

Step 2: Pick the Canonical Format

Step 3: Implement Schema.org Product on Every PDP

This is the universal baseline. Every product detail page should carry a complete Product JSON-LD block:

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "Product",
  "name": "Acme Wireless Headphones X5",
  "image": "https://acme.example/products/x5.jpg",
  "description": "Active-noise-cancelling wireless headphones, 40-hour battery, adaptive EQ.",
  "brand": { "@type": "Brand", "name": "Acme" },
  "sku": "ACX5-BLK",
  "gtin": "0123456789012",
  "offers": {
    "@type": "Offer",
    "url": "https://acme.example/products/x5",
    "priceCurrency": "USD",
    "price": "299.00",
    "availability": "https://schema.org/InStock",
    "itemCondition": "https://schema.org/NewCondition"
  },
  "aggregateRating": {
    "@type": "AggregateRating",
    "ratingValue": "4.6",
    "reviewCount": "1284"
  }
}
</script>

See the Schema.org JSON-LD examples gallery for additional fields (energyConsumption, colour, material, hasMerchantReturnPolicy).

Step 4: Publish a Catalogue Feed

Generate and host a daily-refreshed JSON or CSV catalogue at a stable URL like https://acme.example/catalogue.json. The minimum schema:

{
  "generated_at": "2026-05-15T08:00:00Z",
  "currency": "USD",
  "products": [
    {
      "sku": "ACX5-BLK",
      "gtin": "0123456789012",
      "name": "Acme Wireless Headphones X5",
      "brand": "Acme",
      "category": "headphones / wireless / over-ear",
      "url": "https://acme.example/products/x5",
      "image_url": "https://acme.example/products/x5.jpg",
      "description": "Active-noise-cancelling wireless headphones.",
      "price": 299.00,
      "availability": "in_stock",
      "rating": 4.6,
      "review_count": 1284
    }
  ]
}

Step 5: Expose Through MCP

Add an MCP server with at least search_catalogue, get_product, get_pricing, and check_availability tools. See the MCP server starter template and the how-to-build guide.

Step 6: Validate Across Surfaces

1. Run every PDP through Google's Rich Results test.
2. Submit your Google Merchant Center feed and resolve all warnings.
3. Test a sample of products in ChatGPT Shopping by name and confirm description, price, and availability are correct.
4. Test in Perplexity Shopping and Gemini Shopping similarly.
5. Connect your MCP server to Claude Desktop and run a test query.

Step 7: Keep It Fresh

Stale product data is worse than no data. At minimum:

• Regenerate the catalogue feed daily.
• Set MCP server cache TTL to 1 hour or less for pricing and availability.
• Push immediate updates for price changes, stockouts, and new launches via your Google Merchant Center API.
• Monitor for drift between your storefront and the feed.

Common Mistakes

1. Schema.org Product only on some PDPs. Inconsistent schema is treated as low-trust by AI assistants.
2. Missing price or availability fields. Required for agentic-commerce surfaces.
3. No GTIN or SKU. Disambiguation across multiple retailers requires these.
4. Stale stock status. Selling out-of-stock items via agentic flows damages brand trust fast.
5. No agent-readable returns policy. ChatGPT Shopping and Perplexity Shopping increasingly include returns terms in their answer; if yours isn't published, you default to industry norms.
6. MCP server with no product tools. Brand catalogue is the highest-value MCP tool category in 2026.