Skip to main content

Agent Protocols

How do AI agents coordinate without human intermediaries?

Agent protocols define how autonomous systems discover, communicate, transact, and collaborate. As AI agents become economic actors, they need shared standards — the same way humans needed HTTP for the web and TCP/IP for the internet.

Protocol Stack

LayerProtocolGoverned ByWhat It Enables
CommunicationA2ALinux FoundationAgent discovery, task delegation, coordination
Tool AccessMCPLinux FoundationAI model access to external tools and data
CommerceUCPGoogle (open)Checkout, discounts, fulfillment — agents buying and selling
PaymentAP2Google (open)Payment authorization via verifiable credentials
IdentityVerifiable IntentMastercard + GoogleCryptographic proof linking human intent to agent transaction
Settlementx402, card rails, stablecoinsVariousValue transfer — crypto-native or traditional
AttributionPCPStory ProtocolIP provenance and creative value flow

The Google stack: A2A + UCP + AP2 form a coherent ecosystem — communicate, then commerce, then pay. Each protocol is open (Apache 2.0) and moving toward foundation governance.

The competitor: ACP (Agentic Commerce Protocol) from OpenAI + Stripe takes a different approach — merchant-centric, card-rail native. See Agent Commerce for the standards war analysis.

Commerce Flow

DISCOVER → COMMUNICATE → COMMERCE → AUTHORIZE → SETTLE → EVALUATE
↓ ↓ ↓ ↓ ↓ ↓
Agent A2A UCP AP2 Stablecoin PCP
Cards Messages Checkout VDC or card Reputation
StageProtocolWhat Happens
DiscoverA2A Agent CardsAgent publishes /.well-known/agent.json, clients find it
CommunicateA2A Tasks + MessagesAgents coordinate, delegate, stream results
CommerceUCPAgent browses catalog, builds cart, applies discounts
AuthorizeAP2 + Verifiable IntentAgent proves payment authority; human consent is traceable
Settlex402 / card rails / stablecoinsValue moves from buyer to seller
EvaluatePCP / reputation systemsQuality signal feeds back into discovery

Two Foundations

Both communication protocols are Linux Foundation projects. Complementary, not competing:

A2AMCP
ConnectsAgent to agentAgent to tool
PatternPeer-to-peer coordinationClient-server tool access
Use case"Research this, then delegate analysis""Query this database, read this file"
GovernanceLinux FoundationLinux Foundation

An agent uses MCP to access data sources, then A2A to delegate specialized work to other agents, then UCP/AP2 to buy things on behalf of the user. Same agent, three protocols, one workflow.

Internal Protocols

ProtocolWhat It Governs
PRD HandoffHow the dream team writes PRDs and how engineering reads them — Story Contract, Build Contract (FAVV v2.1), parser detection

Context

  • A2A Protocol — How agents discover and coordinate
  • UCP — How agents buy and sell
  • AP2 — How agents authorize payments
  • MCP — How agents access tools
  • Verifiable Intent — Cryptographic proof linking human intent to agent action
  • Agent Commerce — The standards war for agent transactions
  • Proof of Creativity — IP attribution for AI-generated content
  • Essential Algorithm — The commerce flow IS the essential algorithm: INTENT → ROUTE → INFRA → SETTLE → FEEDBACK
  • Protocols — Algorithms decide the route; protocols enable the handshake
  • Phygital Beings — Agents as economic actors
  • Smart Contracts — On-chain settlement infrastructure

Questions

When agents can discover, communicate, transact, and evaluate each other — who writes the rules of engagement?

  • If Google's stack (A2A + UCP + AP2) and OpenAI's stack (MCP + ACP) both succeed, does the agent that speaks both protocols win?
  • What happens to the concept of "customer experience" when the customer is an algorithm?
  • At what point does protocol governance matter more than protocol design?