Preprint. Request for Comment.

Abstract

Software agents are increasingly able to discover, compose, and invoke one another's capabilities across organizational boundaries. The mechanisms used to discover these agents, however, are diverging. Enterprises, platforms, and developer communities have each adopted different systems, including AI Catalogs, DNS-based discovery such as DNS-AID, enterprise gateways, agent descriptors such as A2A Agent Cards and MCP servers, and a growing set of platform, telecom, and sector-specific registries. Each system is authoritative within its own domain, but no common layer connects them. This fragments the ecosystem into isolated discovery islands and prevents agents from finding and using capabilities across boundaries.\par

We argue that the missing piece is not another discovery protocol but a resolution layer situated between identity and discovery. Given a stable identity or resource identifier, this layer determines which discovery mechanism a requester should use and then defers to that mechanism. Because it reuses the AI Catalog format as its record type, it introduces no new schema. We describe NandaIndex, a concrete realization that acts as a global switchboard across heterogeneous systems, and apply it to three settings: enterprises with mixed discovery infrastructure, small businesses whose agent runtime and agent card are hosted by separate providers, and individuals who do not own a domain. The design extends the existing NANDA and AGNTCY switchboard and identifies the identity, trust, and federation mechanisms that remain to be formalized.

Overview at https://nandaindex.org. Code at https://github.com/projnanda/nanda-index-v2