agentic-product-protocol-mcp

Klarna-style product discovery for AI shopping agents.

Makes product catalogs machine-readable so AI agents can search, compare, and purchase products programmatically — no screen scraping, no landing pages.

Today's e-commerce is built for humans: landing pages, image carousels, "Add to Cart" buttons. AI shopping agents can't efficiently navigate this. They need structured product data — not HTML.

Klarna introduced the Agentic Product Protocol (December 2025) to solve exactly this: a standardized way for merchants to expose their product catalogs to AI agents. Think of it as RSS feeds, but for shopping.

This MCP server implements the core ideas of agentic product discovery:

• Structured search results — not web pages, but clean JSON with name, price, nutrition, ratings
• Product comparison — side-by-side structured comparison across multiple dimensions
• Feed conversion — take any product feed (JSON, CSV, Open Food Facts) and normalize it into an agent-friendly schema
• Schema generation — convert raw product data into the Agentic Product Protocol format
• Availability checking — real-time product status in a machine-readable format

Uses Open Food Facts as a demo data source — works with any product feed.

pip install agentic-product-protocol-mcp

Or with uvx (no install needed):

uvx agentic-product-protocol-mcp

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "product-protocol": {
      "command": "uvx",
      "args": ["agentic-product-protocol-mcp"]
    }
  }
}

claude mcp add product-protocol -- uvx agentic-product-protocol-mcp

| Tool | Description | |------|-------------| | search_products | Search products with structured results (name, nutrition, labels, st