Agent Payments Protocol (AP2)

Google announced AP2 on September 16, 2025, with more than 60 launch partners, including Mastercard, PayPal, Coinbase, and American Express. It is an open specification, developed in the open on GitHub, and is designed to compose with rather than replace existing agent standards: it can be used as an extension of the Agent2Agent (A2A) protocol and the Model Context Protocol (MCP). AP2 answers a narrow but hard question, namely how a merchant can trust that an autonomous agent is genuinely authorized to spend a specific buyer's money on a specific purchase.

The core mechanism is three signed Mandates carried as W3C Verifiable Credentials. The Intent Mandate captures what the user asked for (for example, "find me new white running shoes under a set budget"). The Cart Mandate is signed when the user approves the exact items and price the agent assembled, creating an unchangeable record of what was agreed. The Payment Mandate authorizes the actual charge. Because each Mandate is cryptographically signed, the full sequence from intent to cart to payment produces a non-repudiable audit trail that a merchant, network, or issuer can verify.

AP2 is deliberately payment-agnostic. It treats stablecoins as first-class rails alongside cards and bank transfers, and Google shipped a companion A2A x402 extension for agent-based crypto payments with Coinbase, the Ethereum Foundation, and MetaMask. The intent is a single authorization model that works whether the buyer pays with a card, a bank transfer, or a stablecoin.

AP2 is one layer in a larger agentic-commerce stack, not a single winner. MCP is the tool and context layer that governs what an agent can read and call. The Agentic Commerce Protocol (ACP) from OpenAI and Stripe defines the checkout flow between an agent and a merchant. AP2 sits at the payments and authorization layer, proving what the buyer authorized and what the merchant may charge. These standards are designed to compose, and a single agent-driven purchase can draw on more than one of them.

For a Shopify or ecommerce merchant, AP2 signals that the trust and payment rails for agent-driven buying are being standardized by the largest payment networks, not left to ad hoc integrations. You almost certainly will not implement AP2 yourself; it is plumbing that your payment provider, platform, or the agent ecosystem handles. What you control is whether an agent can find, understand, and trust your catalog in the first place. An authorization protocol only matters once an agent has already chosen to put your product in the cart.

That makes agent-readiness the durable, merchant-side move, independent of which checkout or payment standard wins. Accurate, structured, machine-readable product data (correct price, availability, and variants), strong AI-search visibility, and clean signals like structured data and llms.txt are what get your products surfaced and recommended by buying agents across ChatGPT, Gemini, Perplexity, and Copilot. Naridon focuses on that upstream layer, tracking and fixing whether AI engines cite and recommend your products, rather than acting as a payments or checkout provider.

Illustrative flow: a shopper tells an assistant to buy a specific pair of running shoes under a set budget. The request becomes a signed Intent Mandate. The agent assembles a cart, and when the shopper approves the exact items and total, that approval is captured as a signed Cart Mandate. A Payment Mandate then authorizes the charge. The merchant receives a verifiable record of what was requested, what was agreed, and what may be charged, instead of an opaque bot transaction.