MCP and A2A Are Rails, Not Context

The agent protocol ecosystem has matured rapidly. MCP handles vertical tool integration. A2A handles horizontal agent communication. These protocols are necessary rails — but they carry no context about human expertise. They tell agents how to talk to each other. They do not tell agents how to find humans worth talking about.

What These Protocols Actually Do

MCP, donated by Anthropic to the Linux Foundation in December 2025, standardizes how AI systems connect to tools and data sources. It has 97 million monthly SDK downloads and over 10,000 active servers. Draft proposals define server discovery via structured JSON documents. An IETF draft proposes an mcp:// URI scheme. MCP is vertical: an agent reaching downward to its tools.

A2A, contributed by Google in April 2025, handles horizontal communication. Its core discovery mechanism is the Agent Card — a JSON document at a well-known URL containing identity, capabilities, and authentication requirements. Over 150 organizations participate. IBM's competing ACP protocol merged into A2A in August 2025, consolidating the field.

The Gap: Software Discovery vs Expert Discovery

MCP Server Cards describe what tools a server exposes. A2A Agent Cards describe what an AI agent can do. Both are designed for software entities discovering other software entities. Neither has a way to represent human expertise, verified credentials, or domain authority.

The distinction matters. When an agent needs to call a function, MCP provides the layer. When an agent needs to delegate a task to another agent, A2A handles it. But when a reader's agent needs to find a verified expert on a specific clinical question, neither protocol helps — because the expert is a human with credentials and judgment that exists outside the software ecosystem.

ANS and ANP Are Still Machine-to-Machine

ANS creates a DNS-inspired registry with PKI certificates for identity verification. GoDaddy has built a working ANS registry. ANP uses W3C Decentralized Identifiers. Both extend discovery but remain focused on machine identities. The infrastructure for credential verification exists in fragments — NPI for healthcare, ORCID for researchers, bar directories for lawyers — but no aggregation layer combines these into a trust score that agent protocols can consume.

What an Expert Layer Would Look Like

An Expert Agent Card would extend A2A's concept. It would combine identity (W3C DID), verified credentials (issued by licensing boards), expertise domains, availability, pricing, and communication endpoints. An expert — or an agent acting on their behalf — would publish this card at a well-known URL. The payment rails exist (x402, Lightning L402). The expert endpoints do not. See The Agent Discovery Gap for the full argument.