Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Evolution of AI Agent Registry Solutions: Centralized, Enterprise, and Distributed Approaches

Aditi Singh, Abul Ehtesham, Mahesh Lambe, Jared James Grogan, Abhishek Singh, Saket Kumar, Luca Muscariello, Vijoy Pandey, Guillaume Sauvage De Saint Marc, Pradyumna Chari, Ramesh Raskar

TL;DR

As autonomous AI agents proliferate across domains, traditional web infrastructure struggles with sub-second identity resolution, verifiable metadata, and privacy-preserving discovery. The paper analyzes five registry architectures—MCP Registry, A2A, AGNTCY ADS, Microsoft Entra Agent ID, and NANDA Index—and evaluates them along security, authentication, scalability, and maintainability, highlighting the trade-offs between centralized control, enterprise governance, and distributed resilience. It argues for registries that provide cryptographic identity binding, verifiable metadata, and federated trust, and it outlines design pathways toward an interoperable Internet of AI Agents. The proposed Path Ahead envisions a switchboard-like, federated registry ecosystem with phased deployments to enable cross-domain discovery, policy governance, and verifiable capability semantics at scale.

Abstract

Autonomous AI agents now operate across cloud, enterprise, and decentralized domains, creating demand for registry infrastructures that enable trustworthy discovery, capability negotiation, and identity assurance. We analyze five prominent approaches: (1) MCP Registry (centralized publication of mcp.json descriptors), (2) A2A Agent Cards (decentralized self-describing JSON capability manifests), (3) AGNTCY Agent Directory Service (IPFS Kademlia DHT content routing extended for semantic taxonomy-based content discovery, OCI artifact storage, and Sigstore-backed integrity), (4) Microsoft Entra Agent ID (enterprise SaaS directory with policy and zero-trust integration), and (5) NANDA Index AgentFacts (cryptographically verifiable, privacy-preserving fact model with credentialed assertions). Using four evaluation dimensions: security, authentication, scalability, and maintainability, we surface architectural trade-offs between centralized control, enterprise governance, and distributed resilience. We conclude with design recommendations for an emerging Internet of AI Agents requiring verifiable identity, adaptive discovery flows, and interoperable capability semantics.

Evolution of AI Agent Registry Solutions: Centralized, Enterprise, and Distributed Approaches

TL;DR

As autonomous AI agents proliferate across domains, traditional web infrastructure struggles with sub-second identity resolution, verifiable metadata, and privacy-preserving discovery. The paper analyzes five registry architectures—MCP Registry, A2A, AGNTCY ADS, Microsoft Entra Agent ID, and NANDA Index—and evaluates them along security, authentication, scalability, and maintainability, highlighting the trade-offs between centralized control, enterprise governance, and distributed resilience. It argues for registries that provide cryptographic identity binding, verifiable metadata, and federated trust, and it outlines design pathways toward an interoperable Internet of AI Agents. The proposed Path Ahead envisions a switchboard-like, federated registry ecosystem with phased deployments to enable cross-domain discovery, policy governance, and verifiable capability semantics at scale.

Abstract

Autonomous AI agents now operate across cloud, enterprise, and decentralized domains, creating demand for registry infrastructures that enable trustworthy discovery, capability negotiation, and identity assurance. We analyze five prominent approaches: (1) MCP Registry (centralized publication of mcp.json descriptors), (2) A2A Agent Cards (decentralized self-describing JSON capability manifests), (3) AGNTCY Agent Directory Service (IPFS Kademlia DHT content routing extended for semantic taxonomy-based content discovery, OCI artifact storage, and Sigstore-backed integrity), (4) Microsoft Entra Agent ID (enterprise SaaS directory with policy and zero-trust integration), and (5) NANDA Index AgentFacts (cryptographically verifiable, privacy-preserving fact model with credentialed assertions). Using four evaluation dimensions: security, authentication, scalability, and maintainability, we surface architectural trade-offs between centralized control, enterprise governance, and distributed resilience. We conclude with design recommendations for an emerging Internet of AI Agents requiring verifiable identity, adaptive discovery flows, and interoperable capability semantics.

Paper Structure

This paper contains 34 sections, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Trust Requirements in Agent Registries
  4. Design Evaluation Dimensions
  5. MCP Registry
  6. Security
  7. Authentication
  8. Scalability
  9. Maintenance
  10. Agent2Agent (A2A) Protocol
  11. Security
  12. Authentication
  13. Scalability
  14. Maintainability
  15. AGNTCY Agent Directory Service
  16. ...and 19 more sections

Figures (3)

  • Figure 1: Time-flow (sequence) diagrams for AGNTCY Agent Directory. Vertical lifelines represent: Publisher, Agent Directory API, Local Search, OCI Storage, Global Search (semantic DHT / embedding index), and Content & Remote Server Selection. (a) Publishing: artifact push precedes minimal record publication; sparse semantic (skill $\to$ digest) plus digest $\to$ endpoint mappings are inserted. (b) Discovery: local capability intent resolves via global search to digests, minimal records are fetched, local filtering/ranking occurs, endpoints resolved, and the selected artifact is pulled.
  • Figure 2: Microsoft Entra Agent ID Overview simons2025agentid
  • Figure 3: NANDA Index and AgentFacts Architecture: A modular three-layer system for decentralized AI agent discovery and routing. The Lean Index Layer resolves agent identifiers into signed AgentAddr records containing cryptographic identity, metadata URLs, and routing information, federated across registries. The AgentFacts Layer distributes dynamic, agent-controlled metadata via primary and private URLs, with real-time verifiable credential (VC) status updates for revocation. The Dynamic Resolution Layer enables endpoint resolution through stable, adaptive, and rotating strategies to support privacy, load balancing, and DDoS resilience. This layered design enables scalable, privacy-respecting, and interoperable agent discovery across federated domains raskar2025dnsunlockinginternetai.