Agentic

A developer's guide to autonomous AI-driven commerce, what it is, how agentic payments work, and which protocols power the ecosystem.

What is agentic commerce?

Agentic commerce is an approach to buying and selling in which AI agents act autonomously on behalf of consumers or businesses to research, compare, negotiate, and complete purchases, often without direct human intervention at each step.

Unlike traditional e-commerce, which requires a person to manually search for products, evaluate options, and click through checkout, agentic systems shift that work to software. A user defines a goal and constraints, "book me the cheapest nonstop flight that lands before noon", and the agent executes the full transaction.

The shift is being accelerated by three converging forces: the maturity of large language models that can reason over complex goals, the emergence of standardized agent-to-system protocols (MCP, A2A, UCP, ACP, AP2), and the willingness of major card networks, platforms, and merchants to build agent-aware infrastructure.

Agentic vs. traditional e-commerce

The fundamental shift is not just speed or convenience, it's a change in who is the actor. Traditional e-commerce is designed around human attention and input at every step. Agentic commerce is designed around machine-readable APIs, structured data, and delegated authority.

This means the parts of commerce that used to be UX problems are now becoming protocol problems.

Agentic payments

Agentic payments are financial transactions initiated, authorized, and completed by an AI agent acting on a user's behalf, without the user entering card details, clicking "Pay", or being present at the moment of purchase.

Until recently, this was the missing link. AI agents could handle discovery, comparison, and cart management, but authentication and payment authorization required human intervention. The emergence of dedicated agentic payment frameworks, from card networks, payment platforms, and open standards bodies, has closed this gap.

Three core requirements

Any agentic payment system must solve three fundamental problems:

Card network solutions

The two largest card networks have launched proprietary frameworks:

• Mastercard Agent Pay, uses "Agentic Tokens" to let verified AI agents transact on behalf of verified users, currently piloting with Microsoft Copilot and Google's UCP.
• Visa Intelligent Commerce, built around the Visa Trusted Agent Protocol, currently live with Fiserv and expanding to additional partners.

The protocol landscape

Agentic commerce is powered by a layered stack of complementary protocols. They do not compete, they address different parts of the transaction flow. Understanding the stack is essential for building robust agentic integrations.

A single agentic transaction may use all layers simultaneously: MCP to connect the agent to inventory data, A2A to coordinate with a merchant's own agent, UCP or ACP to structure the checkout flow, and AP2 to securely authorize and settle the payment.

Model Context Protocol (MCP)

MCP is an open protocol that standardizes how LLM applications connect to external tools, data sources, and APIs. Think of it as a universal adapter: instead of writing custom integrations for each AI model and each external system, you build one MCP server and any compatible client can use it.

In the agentic commerce context, MCP is the foundation layer. It allows agents to discover and invoke capabilities, querying inventory, looking up orders, processing refunds, through a standardized JSON-RPC interface over HTTP. It does not define commerce workflows or payment semantics; it provides the infrastructure on top of which those higher-level protocols operate.

Key concepts

• Tools, typed, callable actions the agent can invoke (e.g. search_transactions, perform_refund)
• Resources, structured data the agent can read (product catalogs, order records)
• Prompts, reusable instruction templates exposed by the server

Transport

MCP supports two transport modes: stdio for local processes and Streamable HTTP (or SSE) for remote servers. The Payments Orchestration Platform MCP Server uses Streamable HTTP, which is the standard for production deployments.

Agent-to-Agent Protocol (A2A)

A2A defines how AI agents from different providers communicate with each other, the transport and lifecycle of agent-to-agent calls. While MCP handles how an agent connects to tools and data (agent-to-system), A2A handles how one agent delegates tasks to another agent (agent-to-agent).

In a commerce scenario, a consumer's personal shopping agent might communicate with a retailer's merchant agent via A2A to negotiate pricing, check fulfillment windows, or coordinate multi-item orders across multiple sellers. Each agent can be built by a different company on a different platform, and A2A provides the shared language for them to interoperate.

Role in the stack

A2A sits at the foundation layer alongside MCP. It is considered more of a historical reference today, most newer protocols (UCP, ACP) have absorbed its patterns and built higher-level commerce semantics on top. AP2 is explicitly designed as an extension of A2A, adding payment-specific mandates to agent-to-agent communication.

Agentic Commerce Protocol (ACP)

ACP is a protocol co-developed by OpenAI and Stripe that defines how AI agents interact with merchant commerce systems to initiate and complete purchases. It is the protocol underlying ChatGPT's shopping checkout features, which launched in September 2025 with Etsy and over 1 million Shopify merchants.

ACP's primary focus is the checkout layer. It enables agents to initiate transactions using a merchant's existing payment infrastructure through Shared Payment Tokens, secure representations of the user's payment credentials that can be passed without exposing raw card data. Businesses publish their checkout configuration via standard APIs or via MCP, making their products purchasable through any ACP-compatible agent.

Scope

ACP covers physical and digital goods, subscriptions, and asynchronous purchases. It is intentionally focused: it solves the checkout and payment moment rather than the full commerce lifecycle. For businesses targeting OpenAI's ecosystem (ChatGPT and its successors), ACP is the primary integration target.

Relationship to MCP

Merchants can publish their ACP checkout configuration via MCP servers, making it discoverable to any MCP-compatible agent, not just ChatGPT. This means ACP and MCP are complementary: MCP provides the discovery and tool infrastructure, ACP provides the checkout semantics.

Agent Payments Protocol (AP2)

AP2 is an open protocol developed by Google with more than 60 collaborating organizations, including Mastercard, Visa, PayPal, American Express, Adyen, and Coinbase, to securely authorize and settle agent-initiated payments. It operates at the payment layer, one level above MCP and A2A.

The mandate model

AP2's core innovation is the mandate: a tamper-evident, cryptographically signed digital contract that defines exactly what an agent is authorized to do on a customer's behalf. Mandates use the W3C Verifiable Credentials standard and are signed using ECDSA P-256 keys.

There are three mandate types that create an auditable trail across the transaction lifecycle:

Human-present vs. human-not-present

AP2 distinguishes between two transaction modes: human-present (HP), where the user is available to approve at the moment of purchase, and human-not-present (HNP), where the agent acts on a pre-signed mandate without real-time user interaction. This maps to SCA (Strong Customer Authentication) requirements under PSD2 in Europe.

Universal Commerce Protocol (UCP)

UCP is an open standard co-developed by Google and Shopify, with Etsy, Wayfair, Target, and Walmart as founding contributors, and endorsed by 20+ global partners including Adyen, American Express, Mastercard, Stripe, Visa, and Zalando. It is the most comprehensive of the agentic commerce protocols, covering the entire commerce journey from discovery to post-purchase support.

What UCP solves

Traditional commerce infrastructure creates an N×N integration problem: every merchant must build a custom connection to every AI platform or sales channel. UCP collapses this into a single integration point, merchants expose their capabilities once, and any UCP-compatible agent can interact with them.

Scope

Unlike ACP, which focuses on the checkout moment, UCP standardizes the entire commerce lifecycle: product discovery, cart management, checkout, discount application, loyalty programs, order management, fulfillment, and returns. Each functional area is versioned independently so the core stays stable as new capabilities are added.

Architecture

UCP uses a capability discovery model: merchants publish a profile at /.well-known/ucp declaring what they support. Agents fetch this profile, compute the intersection with their own capabilities, and negotiate what the interaction can include. This is analogous to HTTP content negotiation, applied to commerce.

Transport and interoperability

UCP supports REST APIs, MCP bindings, and A2A communication, merchants choose the integration method that fits their existing stack. It is fully compatible with AP2 for payment authorization and with existing card network frameworks (Visa Intelligent Commerce, Mastercard Agent Pay).

Protocol comparison

Where Payments Orchestration Platform fits

Payments Orchestration Platform's MCP Server sits at the foundation layer of the agentic commerce stack. By exposing payment operations as MCP tools, Payments Orchestration Platform enables any LLM client or orchestration framework to discover and invoke payment capabilities, including capture, refund, chargeback management, and transaction reporting, through a standardized interface.

As the protocol ecosystem matures, Payments Orchestration Platform's MCP-based integration provides a stable foundation that higher-level commerce and payment protocols, UCP, ACP, AP2, can build upon. Merchants integrating with Payments Orchestration Platform today are positioned to extend into the full agentic commerce stack as those standards reach production readiness.

Next steps

• Set up the Payments Orchestration Platform MCP Server, see the Integration Tutorial
• Review the MCP specification for advanced client configuration