Agent-to-agent commerce: A2A, AP2, MCP, and the receipt that holds it together

Three protocols are converging into the wire format for agent-to-agent commerce: Google's A2A, the Agent Payments Protocol (AP2), and the Model Context Protocol (MCP). Each solves one slice of the problem. None of them solves evidence. This page maps how Hive's post-quantum receipt envelope sits underneath all three.

What "agent-to-agent commerce" actually means

An autonomous agent is software that holds budget, takes initiative, and transacts with other agents on behalf of a principal. The principal might be a human, a company, or a regulator. When two agents transact — agent A asks agent B for a service, agent B delivers, value moves — every party in the chain needs evidence that the transaction happened, that it was authorized, and that the participants were who they claimed to be.

This is harder than it sounds because the agent population is heterogeneous. One agent runs on Google Vertex with A2A. Another exposes itself as an MCP tool over Streamable HTTP. A third lives behind an AP2 endpoint and quotes prices in USDC. The protocols are not interoperable at the evidence layer.

A2A — Google's agent-to-agent protocol

A2A is Google's open protocol for autonomous agents to discover each other, negotiate capabilities, and exchange messages. It defines an Agent Card schema (a JSON document an agent publishes describing what it can do), a Task envelope, and a long-running streaming session. A2A handles capability discovery, negotiation, and message exchange — but it does not specify how evidence of a completed task is signed, stored, or verified.

An A2A task can complete successfully and disappear. There is no canonical, post-quantum-signed receipt that survives the session. That is the gap Hive fills.

AP2 — Agent Payments Protocol

AP2 is the agent payments specification — value flow between agents. It defines payment intents, quotes, authorizations, and settlements. AP2 settles in stablecoin (USDC is the dominant rail) and is increasingly the wire format for autonomous-agent budgets. AP2 produces a settlement record on-chain but the on-chain record is not a substitute for an evidence envelope: it doesn't carry the agent identities, the negotiation context, the model versions, or any post-quantum signature surface.

Hive bolts a post-quantum receipt onto every AP2 settlement: settlement_ref in the Hive receipt is the AP2 transaction hash; sig_dsa binds the negotiation context, agent DIDs, and model versions to that settlement.

MCP — Model Context Protocol

MCP is the protocol that lets an LLM-backed agent discover and call tools at inference time. A tool is an HTTP endpoint that exposes a JSON-RPC 2.0 surface and a manifest describing its capabilities. MCP has rapidly become the de-facto integration layer between models and external services.

An MCP tool call is an interaction worth receipting: the agent asked for X, the tool produced Y, the tool charged Z. Hive ships a family of MCP servers that wrap our endpoints, and every MCP call through the Hive gateway can produce a post-quantum receipt as part of the response envelope.

The missing layer: evidence

None of A2A, AP2, or MCP specifies a post-quantum signed evidence envelope. Each assumes that the transport layer's TLS plus an application-layer signature is enough. For regulated AI workloads it is not — the threat model has to include a quantum-capable adversary running harvest-now-decrypt-later, and it has to include the auditor who shows up in 2032 wanting to verify a 2026 transaction.

An evidence envelope has to do four things: (1) bind the parties' identities cryptographically, (2) bind the transaction context cryptographically, (3) survive primitive deprecation, (4) be verifiable offline. Hive's receipt envelope does all four.

How Hive receipts unify the three

The Hive receipt is protocol-agnostic. The same envelope shape sits under A2A tasks, AP2 settlements, and MCP tool calls. The field guide has the full envelope spec; the short version is:

• Same canonical CBOR encoding regardless of transport protocol.
• Same dual signature (Ed25519 + ML-DSA-65) regardless of whether the call originated on A2A, AP2, or MCP.
• Same settlement_ref field linking to the on-chain USDC settlement on Base 8453.
• Same offline reference verifier.

This means an organization that adopts Hive as its evidence layer can let its agents speak any of the three wire protocols without fragmenting the audit trail.

Integration patterns

Three patterns cover most deployments:

The developer documentation walks through each pattern with example code.