PKPProduct Knowledge Protocol

Open standard for AI agent product knowledge

Get Started

View on GitHub

Web-native

Data lives in /.well-known/pkp/ on vendor domains. No central database required.

Agent-first

Designed for AI agents to discover, compare, and recommend products intelligently.

Explicit Confidence

Every data block has source attribution and verification timestamps.

Comparable by Design

Category-specific schemas ensure products can be meaningfully compared.

What is PKP?

PKP (Product Knowledge Protocol) is an open format + category schemas for representing product knowledge in a way that any AI agent can consume.

It's the missing layer in the agentic commerce stack - solving "what to buy and why" before payment (AP2) and transaction (UCP) protocols.

yaml

# Example PRODUCT.md
---
schema: pkp/1.0
sku: "galaxy-s25-ultra"
brand: "Samsung"
name: "Galaxy S25 Ultra"
category: "celulares/smartphones"
summary: "Flagship smartphone with S Pen and 200MP camera"

specs:
  display_size: 6.9
  processor: "Snapdragon 8 Elite"
  ram_gb: 12
  storage_gb: 256
  main_camera_mp: 200
  battery_mah: 5000

confidence:
  specs:
    source: manufacturer
    verified_at: "2025-01-15"
---

## Overview
The Galaxy S25 Ultra is Samsung's most advanced smartphone...

Quick Install

bash

# Install the CLI
npm install -g pkp

# Initialize a new catalog
pkp init my-catalog

# Build for deployment
pkp build my-catalog

Packages

Package	Description
`pkp`	CLI for validating, building, and generating product data
`@pkprotocol/spec`	Schemas, parser, and validator
`@pkprotocol/catalog-server`	MCP server for serving catalog data
`@pkprotocol/registry-server`	Central discovery and indexing service
`@pkprotocol/skills`	AI prompts for product assistants