Local Agent Discovery for A2A (LAD-A2A)
An open protocol for discovering Agent-to-Agent capable AI agents on a local network, such as a hotel, office or hospital, using mDNS and well-known endpoints, with signed AgentCards and explicit user consent before any connection is trusted.
LAD-A2A fills a gap it identifies in the AI agent ecosystem: when a device joins a network, its AI assistant has no standard way to discover local agents, such as a hotel's concierge agent or a hospital's wayfinding agent. It defines zero-configuration discovery over mDNS and DNS-SD well-known endpoints, and separates channel authentication, where TLS only proves the assistant reached the host named in a URL, from identity verification, which anchors a domain, JWS or DID to an organization, since TLS to an attacker-controlled host proves nothing about who that host actually is.
The protocol requires explicit human approval before first contact with a newly discovered agent, keyed to its verified identity rather than auto-approved, and hands off to the standard A2A protocol for the actual agent-to-agent communication once an agent is verified and approved; it positions itself as the discovery and trust layer beneath A2A, the same way MCP handles agent-to-tool integration.