Ayodele Aransiola

Back to Blog

A2A, MCP, and AP2: The Protocol Stack Powering Agentic AI Commerce

September 26, 2025 By Ayodele
A2A
MCP
AP2
AI Agents
Artificial Intelligence

AI is entering a new era, one where autonomous agents don’t just respond to prompts but can act on behalf of users, collaborate with other agents, and even conduct secure financial transactions. To make this possible, we need more than powerful models. We need protocols that allow agents to interoperate, share context, and transact responsibly.

That’s where three emerging standards come in:

  • A2A (Agent-to-Agent Protocol) – agents collaborating with each other.
  • MCP (Model Context Protocol) – agents accessing tools, APIs, and data.
  • AP2 (Agent Payments Protocol) – agents making payments under user-controlled mandates.

Together, they form the backbone of agentic commerce.

Let's go a little deeper, explaining each pillar:

A2A: Agent-to-Agent Collaboration

The Agent-to-Agent (A2A) protocol defines how independent AI agents talk to each other.

  • Agents can delegate tasks: e.g., a travel-booking agent asks a payment agent to prepare checkout.
  • They can negotiate outcomes: e.g., a sourcing agent and supplier agent exchange quotes.
  • They remain secure and modular: each agent runs independently, exposing only what’s necessary.

Think of A2A as the “HTTP for agents”: that is, a common protocol that lets them coordinate, regardless of who built them.

MCP: Model Context Protocol

The Model Context Protocol (MCP) (pioneered by Anthropic, now supported broadly) ensures agents have structured ways to access context and tools.

  • An agent can fetch user files, calendar data, or enterprise APIs.
  • It standardizes context exchange so agents don’t need custom integrations for every tool.
  • MCP acts like a universal adapter, much like USB-C, but for AI agents.

Example: A scheduling agent using MCP could connect to Google Calendar, Outlook, and internal company APIs without hard-coding integrations.

AP2: Agent Payments Protocol

The newest piece is AP2 (Agent Payments Protocol), proposed by Google and partners. Traditionally, payments assume a human presses “Pay Now.” But if agents are acting independently, we need a way to:

  • Capture the user’s intent.
  • Authorize the agent to act within limits.
  • Ensure transactions are verifiable and auditable.

That’s what AP2 introduces, via Mandates:

  • Intent Mandate – “Buy me two concert tickets, best available seats, under $300.”
  • Cart Mandate: “This is the specific order your agent prepared. Approve?”

With these cryptographically signed mandates, an agent can pay securely on your behalf, while merchants and payment providers know the purchase is authorized.

How do They Work Together?

Let’s walk through an example:

  1. User instruction: “Find me the best flight to San Francisco under $800 and book it.”
  2. A2A in action: The travel agent talks to an airline agent and a payment agent to coordinate.
  3. MCP in action: The agent accesses flight APIs, loyalty points data, and seat maps through MCP.
  4. AP2 in action: Once an option is found, the agent generates a Cart Mandate. The user approves it. The payment agent completes the transaction securely.

Why This Matters

This protocol trio unlocks:

  • Interoperability: Agents from different products or providers can work together.
  • Trust & Security: Mandates ensure payments aren’t rogue.
  • Scale: Businesses and users can safely delegate tasks to AI.

Conclusion

AI agents are moving beyond experimental projects into real-world solutions.