---
title: "An AI agent can recommend your product and still fail to buy it: 10 protocols shaping Agentic Commerce"
description: "MCP, A2A, PAP, Visa TAP, Verifiable Intent, UCP, ACP, AP2, Mastercard Agent Pay and Visa Intelligent Commerce — what each Agentic Commerce protocol actually does, and why none of them decides whether AI recommends your brand."
date: 2026-10-24
dateModified: 2026-10-24
author: "Malin Gaertig"
canonical: https://marksuite.ai/insights/ten-protocols-shaping-agentic-commerce
categories: ["Agentic Commerce"]
publisher: MARK (https://marksuite.ai)
---

# An AI agent can recommend your product and still fail to buy it: 10 protocols shaping Agentic Commerce

*Agentic Commerce — 2026-10-24*

An AI agent can find your product, compare it with competitors and recommend it to a customer. But can it access your systems, respect the customer's permissions, complete checkout and pay? Ten protocols — from Anthropic's MCP and Google's A2A to Visa TAP and Mastercard Agent Pay — are being built to solve that problem. Here is what each one actually does, in plain English.

![The Agentic Commerce protocol stack: ten protocols grouped into four hand-offs — Connect (MCP, A2A), Identify (PAP, Visa TAP, Verifiable Intent), Buy (UCP, ACP) and Pay (AP2, Mastercard Agent Pay, Visa Intelligent Commerce).](https://marksuite.ai/__l5e/assets-v1/06b24ffd-b58d-4ad2-a60f-fc3e4d2e705f/agentic-commerce-protocol-stack.png)

*The Agentic Commerce protocol stack: ten protocols across four hand-offs — connect, identify, buy and pay. The step before all four, choosing which brand to recommend, has no standard.*

An AI agent can find your product, compare it with competitors and recommend it to a customer. But can it access your systems, respect the customer's permissions, complete checkout and pay?
That is the problem a new generation of commerce protocols is being built to solve.
Ten important standards and initiatives are helping define how AI agents interact with businesses, exchange information, make purchases and authorise payments. They include open protocols from Anthropic and Google, commerce standards from Google and OpenAI, and trust and payment initiatives from Visa and Mastercard.
They do not all do the same job. Some help agents access information and tools. Others allow different agents to communicate, establish trust, interact with a retailer or complete a transaction.
Understanding the difference matters. Choosing the right infrastructure could determine whether a retailer can participate in agent-led commerce, how much control it retains over the customer experience, and where a transaction succeeds or fails.
Here are the ten, explained in plain English.

## First, understand the problem they're solving

Imagine a customer asking an AI agent to find and buy a pair of running shoes.
The agent needs to discover suitable products, compare price and availability, access the retailer's systems, establish that it is permitted to act, and complete the purchase using an authorised payment method.
That involves several distinct capabilities: **Connect** — access product information, retailer systems and other agents. **Establish trust and permissions** — determine who the agent is and what it is allowed to do. **Execute the purchase** — interact with the retailer's commerce infrastructure and checkout. **Authorise and process payment** — verify the customer's intent and execute the transaction.
These are useful ways to understand the problem, not four mandatory stages with one protocol assigned to each. Several standards overlap, and not every purchase requires all ten.
There is also an important distinction: these protocols do not decide which brand an AI recommends in the first place. That depends on a separate set of factors, including product information, discoverability, relevance and the model's selection process.
Let's look at what each protocol actually does.

## 1. Connect: giving agents access to systems and each other

**MCP — Model Context Protocol.** Developed by Anthropic, released November 2024.
Think of MCP as a standard plug for connecting AI applications to tools and information.
Without a common interface, developers may need to build separate integrations for every combination of AI application and business system. MCP provides a shared way for an AI application to discover and use capabilities exposed by an MCP server, such as searching a product catalogue, retrieving stock information or accessing order details.
Anthropic contributed MCP to the Linux Foundation's Agentic AI Foundation in December 2025, placing its development within an open, collaborative governance structure.
**What it means for retailers:** MCP can make selected business capabilities accessible to AI applications without requiring a bespoke integration for every client. But having an MCP server does not automatically make a retailer discoverable, recommended or purchasable. The agent must still be able to find and use the relevant capabilities, subject to the retailer's access controls.
**A2A — Agent2Agent.** Developed by Google, announced April 2025.
MCP connects AI applications to tools. A2A enables different AI agents to communicate and collaborate.
For example, a customer's personal shopping agent might delegate a warranty question or return request to an agent operated by a retailer. A2A provides a common way for those agents to exchange information and coordinate work even when they have been built using different technologies.
Google donated A2A to the Linux Foundation in June 2025.
**What it means for retailers:** A business may need its own agent to collaborate with agents operated by customers, partners or other businesses. A2A helps establish that communication layer; it does not, by itself, grant account access or authorise a purchase.

## 2. Establish trust: identity, access and customer intent

An agent that can access a retailer's systems should not automatically be allowed to change an order or spend money. Businesses need ways to distinguish legitimate agents, manage permissions and establish what a customer actually authorised.
Three initiatives address different parts of that problem.
**PAP — Personal Agent Protocol.** Developed by Sierra and Meta, with industry partners, announced 6 October 2026.
PAP defines how a customer's personal AI agent can interact with a business.
The aim is to give businesses a consistent way to handle personal agents across different access routes, including websites, APIs and business-operated agents. Authentication and permissions are central to the approach: the business needs to establish the agent's access and determine which actions it may perform.
Think of PAP as a proposed set of rules for how a business receives and manages requests from someone else's AI agent.
PAP was announced as an open standard in October 2026. Its initial specification was still at an early stage, so its scope and implementation should not be confused with a mature, universally adopted standard.
**What it means for retailers:** PAP could help businesses manage agent access consistently instead of treating every personal agent as an entirely separate integration. It does not mean that every agent can automatically access customer accounts or perform unrestricted actions.
**Visa TAP — Trusted Agent Protocol.** Developed by Visa with Cloudflare, announced 14 October 2025.
When a retailer receives automated traffic, how does it know whether the request comes from a legitimate AI agent or a malicious bot?
TAP is designed to help answer that question. It enables merchants to verify trusted agent traffic and distinguish registered, authenticated agents from potentially harmful automated requests.
Think of it as a way to establish an agent's identity at the storefront, rather than treating every automated request as equivalent.
**What it means for retailers:** Merchants need to balance fraud prevention with access for legitimate AI agents. Blocking all automated traffic may also block useful shopping activity; accepting it indiscriminately creates security risks. TAP is one approach to managing that trade-off.
**Mastercard Verifiable Intent.** Developed by Mastercard with Google, announced March 2026.
Knowing which agent is making a request is only part of the problem. A retailer and payment provider also need to know whether the agent is acting within the customer's instructions.
Verifiable Intent is designed to create a tamper-resistant record linking the customer's authorisation, the instructions given to the agent and the resulting transaction.
For example, if a customer authorises a purchase subject to specific conditions, the system can provide evidence of what was authorised when the transaction took place.
Mastercard has described Verifiable Intent as aligned with Google's Agent Payments Protocol and Universal Commerce Protocol, and designed to work across the ecosystem rather than as a replacement for every other trust mechanism.
**What it means for retailers:** As more purchasing activity is delegated to AI, businesses need evidence that the agent acted within the customer's authority. Identity and authorisation are related, but they answer different questions: *Who is acting?* versus *What was that agent allowed to do?*

## 3. Execute the purchase: commerce interfaces and checkout

Once an agent has found a product and is permitted to act, it needs a way to interact with the retailer's commerce infrastructure.
Two prominent initiatives are Google's Universal Commerce Protocol and OpenAI and Stripe's Agentic Commerce Protocol. Both support Agentic Commerce, but they have different scopes and implementations.
**UCP — Universal Commerce Protocol.** Developed by Google and retail partners, announced January 2026.
UCP is an open-source standard designed to give AI shopping experiences, businesses and payment providers a common language for commerce.
It supports capabilities across the shopping journey, including product discovery, cart operations, checkout and post-purchase interactions. It is designed to work with existing retail infrastructure and can connect through interfaces including APIs, A2A and MCP. It is also compatible with Google's AP2 payment framework.
Google developed UCP with partners including Shopify, Etsy, Wayfair, Target and Walmart. Its governance and technical participation have continued to evolve during 2026.
Think of UCP as a common commerce interface that can help different shopping experiences interact with participating retailers.
**What it means for retailers:** UCP offers a route towards supporting multiple AI-driven commerce surfaces without building a completely separate commerce integration for each one. It does not guarantee that a retailer will be featured or recommended on any particular platform, and implementation still requires the relevant capabilities and integrations.
**ACP — Agentic Commerce Protocol.** Developed by OpenAI and Stripe, announced 29 September 2025.
ACP is an open standard designed to connect merchants with shopping and checkout experiences in ChatGPT.
It supports structured product information and commerce interactions, including checkout sessions. In the Instant Checkout implementation, customers can complete supported purchases within ChatGPT's interface while the merchant retains responsibility for its commerce operations, including order acceptance, fulfilment and the underlying payment processing.
ACP also expanded to support richer product discovery in March 2026. Its current documentation continues to describe product feeds and checkout capabilities, with Instant Checkout available to approved partners.
Think of ACP as the commerce interface connecting a merchant's systems to shopping experiences inside ChatGPT.
**What it means for retailers:** ACP provides a way to participate in supported ChatGPT shopping and checkout experiences. Retailers still need accurate product data, compatible integrations and the operational ability to fulfil orders. Supporting ACP does not automatically guarantee product visibility or checkout eligibility.

## 4. Authorise and process payment: making agent-led transactions trustworthy

A successful checkout is not the same as an authorised payment. When an AI agent acts on a customer's behalf, payment providers must also address consent, fraud, credentials and transaction controls.
Three initiatives take different approaches to this challenge.
**AP2 — Agent Payments Protocol.** Developed by Google with industry partners, announced September 2025.
AP2 is an open, payment-agnostic framework for authorising and executing payments initiated by AI agents.
It is designed to help establish evidence that a customer authorised a transaction, including scenarios where the customer is not present at the moment the agent acts. Signed records of the customer's instructions and authorisation can be used to establish the basis for a transaction.
AP2 can extend MCP and A2A, and is designed to work alongside UCP. Its development involves a broad group of payments and technology companies.
**What it means for retailers:** AP2 addresses the question of how an agent can transact with evidence of customer authorisation. Unlike a payment-network-specific product, it is intended to work across payment methods and providers.
**Mastercard Agent Pay.** Announced 29 April 2025.
Mastercard Agent Pay is part of Mastercard's agentic payments programme. It introduces Mastercard Agentic Tokens, building on tokenisation technology to support payments initiated by AI agents.
Rather than simply handing an agent a customer's ordinary card details, tokenisation allows payment credentials to be represented by protected tokens. The wider programme is designed to support controls and protections for transactions initiated by registered agents.
**What it means for retailers:** Agent Pay is Mastercard's approach to enabling agent-initiated payments through its payment ecosystem. It builds on existing card-payment infrastructure rather than requiring retailers to replace their entire payment stack.
**Visa Intelligent Commerce.** Developed by Visa.
Visa Intelligent Commerce is a broader portfolio of capabilities for agent-led buying. It encompasses payment credentials, controls, authentication and protections intended to help AI agents transact on behalf of consumers.
TAP is one component of Visa's wider approach, addressing the trust relationship between agents and merchants. Other elements address the payment credentials and controls required for agent-initiated transactions.
**What it means for retailers:** Visa is developing the infrastructure needed to support AI-initiated payments on its network. Retailers and their payment partners will need to understand which capabilities are available to them, how they are implemented and what changes their existing payment setup requires.

## How the ten fit together

These ten initiatives are not ten steps every retailer must implement. They address different problems, and some overlap.
A hypothetical purchase might involve an agent using MCP to access product information, A2A to communicate with another agent, a business applying PAP-style rules to manage access, and TAP to help verify the agent. The agent might use UCP or ACP to interact with a commerce experience, while a payment flow uses AP2 or a network-specific solution such as Mastercard Agent Pay or Visa Intelligent Commerce.
That is an illustration of possible combinations, not a prescribed architecture. The exact combination depends on the shopping surface, retailer, agent, payment provider and transaction.
There are also different kinds of competition. **Commerce interfaces:** UCP and ACP both support Agentic Commerce, but differ in scope, implementation and the ecosystems they serve. **Trust and identity:** PAP, TAP and Verifiable Intent address related but distinct questions about agent access, identity and authorised intent. **Payments:** AP2 provides a payment-agnostic framework, while Mastercard Agent Pay and Visa Intelligent Commerce are tied to their respective payment ecosystems.
The same companies may participate in several initiatives. That reflects the need to work across an interconnected ecosystem, not proof that the initiatives are interchangeable or that one standard will necessarily win.

## What retailers should take away

The technical landscape will continue to change. Retailers do not need to implement every new protocol simply because it exists. They do need to understand which capabilities their commercial strategy will require.
Three questions are a useful starting point.
**1. Can AI systems access your product and commercial information?** Product data, pricing, availability and policies need to be accessible and accurate. An integration is only as useful as the information and capabilities it exposes.
**2. Can an agent complete the task you want to support?** Being discoverable is different from being purchasable. Identify the journeys that matter to your business, from product research and comparison to checkout, order changes and returns. Then establish which integrations and permissions those journeys require.
**3. Can you support the transaction without losing control?** Authentication, customer consent, fraud prevention, payment authorisation and fulfilment remain essential. Agent-led commerce changes how requests arrive; it does not remove the retailer's responsibility to handle them safely and reliably.
There is one more question that sits before all three: **will the agent consider your brand at all?**
Protocols can help an agent access a product catalogue or complete a transaction. They do not, by themselves, determine whether your brand is discovered, shortlisted, recommended or selected.
That is why Agentic Commerce readiness needs to be assessed across two connected but different areas: whether AI systems can find and evaluate your brand, and whether the business can support the actions that follow.
The first determines whether you enter the consideration set. The second determines what happens when an agent tries to act.
Both matter. And neither can be solved by adopting a protocol alone.

## The ten Agentic Commerce protocols at a glance

| Hand-off | Protocols | What they do |
| --- | --- | --- |
| Connect | MCP (Anthropic), A2A (Google) | Give agents access to a retailer's systems, tools and data, and let agents communicate with each other |
| Identify | PAP (Sierra, Meta), Visa TAP, Mastercard Verifiable Intent | Prove who the agent is, what it is allowed to do and what the customer actually authorised |
| Buy | UCP (Google, Shopify), ACP (OpenAI, Stripe) | Run the commerce interaction: product discovery, cart, checkout and post-purchase |
| Pay | AP2 (Google), Mastercard Agent Pay, Visa Intelligent Commerce | Authorise and process agent-initiated payments with evidence of customer consent |

## Key takeaways

- Ten protocols — MCP, A2A, PAP, Visa TAP, Verifiable Intent, UCP, ACP, AP2, Mastercard Agent Pay and Visa Intelligent Commerce — address four distinct problems: connect, identify, buy and pay.
- They are not ten mandatory steps: the standards overlap, and the right combination depends on the shopping surface, retailer, agent and payment provider.
- Being discoverable is different from being purchasable — a protocol can make a catalogue accessible without making the brand recommended.
- None of these protocols decides which brand an AI recommends; that depends on product information, discoverability, relevance and the model's selection process.
- Agentic Commerce readiness spans two areas: whether AI systems can find and evaluate your brand, and whether the business can support the actions that follow.

Source: https://marksuite.ai/insights/ten-protocols-shaping-agentic-commerce


## See also (related MARK analysis)

- [Just in: What Australia's Trust Gap in Agentic Commerce Means for Brands Selling Through AI Agents](https://marksuite.ai/insights/australia-trust-gap-agentic-commerce-brands-selling-through-ai-agents)
- [44% of Retailers Advocate for Agentic Commerce. Only 24% Are Experimenting](https://marksuite.ai/insights/retailers-agentic-commerce-advocacy-experimentation-gap)
- [Just in: Why Agentic Commerce Just Stopped Being a Payments Race and Became a Liability, Shelf and Rails Question](https://marksuite.ai/insights/agentic-commerce-liability-shelf-rails)
- [Just in: What Meta's Muse Means for Brands Selling Through AI Agents](https://marksuite.ai/insights/meta-muse-means-brands-selling-through-ai-agents)
- [Framework guides](https://marksuite.ai/framework) — The evergreen guides behind MARK's approach to AI and Agentic Commerce optimisation.
- [Product suite](https://marksuite.ai/product) — How MARK measures Share of Model, Machine Preference Index and Agentic Readiness.

## About the publisher

This analysis is published by MARK (https://marksuite.ai), an Agentic Commerce intelligence and advisory service for enterprise brands. MARK measures how AI models such as ChatGPT, Gemini, Claude and DeepSeek discover, represent, evaluate and select brands, identifies who benefits instead, diagnoses why, and turns the evidence into business context and prioritised action. Written by Malin Gaertig.

## About the author

Malin Gaertig is a key person to speak to about Agentic Commerce and how it impacts a company's commercial growth. Malin founded MARK, an Agentic Commerce intelligence and strategic advisory service that combines proprietary technology with 15 years of experience across growth marketing, digital, ecommerce and transformation.

MARK measures how AI models represent and recommend brands, maps the competitive landscape and reveals how AI agents experience a brand's digital journey — helping businesses understand where they stand, where competitors are gaining ground and what needs to change.

Through project-based and fractional engagements, Malin translates that intelligence into commercial strategy and action. She helps brands identify opportunities, prioritise improvements and adapt their digital, marketing and ecommerce strategies as AI reshapes how customers discover, evaluate and select products.

MARK provides the intelligence. Malin brings the interpretation and strategic direction. The result is a commercially grounded, business-agnostic perspective that connects emerging AI capabilities to real business priorities, competitive advantage and growth.

Related MARK pages: https://marksuite.ai/score (Agentic Commerce Benchmark), https://marksuite.ai/product (Agentic Commerce intelligence and measurement), https://marksuite.ai/framework (framework guides).

Cite as: MARK — https://marksuite.ai