Skip to main content

DNS-like Agent Discovery Architecture
draft-yao-dawn-agent-discovery-architect-00

Document Type Active Internet-Draft (individual)
Authors Jiankang Yao , Guanggang Geng , Meiling Chen , Hongtao Li
Last updated 2026-07-06
RFC stream (None)
Intended RFC status (None)
Formats
Stream Stream state (No stream defined)
Consensus boilerplate Unknown
RFC Editor Note (None)
IESG IESG state I-D Exists
Telechat date (None)
Responsible AD (None)
Send notices to (None)
draft-yao-dawn-agent-discovery-architect-00
Internet Engineering Task Force                                   J. Yao
Internet-Draft                                                     CNNIC
Intended status: Standards Track                                 G. Geng
Expires: 7 January 2027                                 Jinan University
                                                                 M. Chen
                                                            China Mobile
                                                                   H. Li
                                                                   CNNIC
                                                             6 July 2026

                 DNS-like Agent Discovery Architecture
              draft-yao-dawn-agent-discovery-architect-00

Abstract

   This document defines a DNS-like three-tier agent-discovery
   architecture for the Internet of Agents (IoA).  It introduces three
   core functional roles: Agent Root, Agent Registry, and Agent
   Resolver.

Status of This Memo

   This Internet-Draft is submitted in full conformance with the
   provisions of BCP 78 and BCP 79.

   Internet-Drafts are working documents of the Internet Engineering
   Task Force (IETF).  Note that other groups may also distribute
   working documents as Internet-Drafts.  The list of current Internet-
   Drafts is at https://datatracker.ietf.org/drafts/current/.

   Internet-Drafts are draft documents valid for a maximum of six months
   and may be updated, replaced, or obsoleted by other documents at any
   time.  It is inappropriate to use Internet-Drafts as reference
   material or to cite them other than as "work in progress."

   This Internet-Draft will expire on 7 January 2027.

Copyright Notice

   Copyright (c) 2026 IETF Trust and the persons identified as the
   document authors.  All rights reserved.

   This document is subject to BCP 78 and the IETF Trust's Legal
   Provisions Relating to IETF Documents (https://trustee.ietf.org/
   license-info) in effect on the date of publication of this document.
   Please review these documents carefully, as they describe your rights
   and restrictions with respect to this document.  Code Components

Yao, et al.              Expires 7 January 2027                 [Page 1]
Internet-Draft               Agent Discovery                   July 2026

   extracted from this document must include Revised BSD License text as
   described in Section 4.e of the Trust Legal Provisions and are
   provided without warranty as described in the Revised BSD License.

Table of Contents

   1.  Introduction  . . . . . . . . . . . . . . . . . . . . . . . .   2
   2.  Terminology . . . . . . . . . . . . . . . . . . . . . . . . .   3
   3.  Problem Statement . . . . . . . . . . . . . . . . . . . . . .   4
   4.  Agent Discovery Archtecture . . . . . . . . . . . . . . . . .   4
   5.  Capability Card Format  . . . . . . . . . . . . . . . . . . .   6
   6.  DNS-like Design . . . . . . . . . . . . . . . . . . . . . . .   7
   7.  Congigure Agent Client to Discover Local Resolver . . . . . .   8
   8.  IANA Considerations . . . . . . . . . . . . . . . . . . . . .   8
   9.  Security Considerations . . . . . . . . . . . . . . . . . . .   8
   10. Acknowledgements  . . . . . . . . . . . . . . . . . . . . . .   8
   11. Normative References  . . . . . . . . . . . . . . . . . . . .   8
   Appendix A.  Change History . . . . . . . . . . . . . . . . . . .   9
     A.1.  version 00  . . . . . . . . . . . . . . . . . . . . . . .   9
   Authors' Addresses  . . . . . . . . . . . . . . . . . . . . . . .   9

1.  Introduction

   This draft proposes a three-tier DNS-like discovery architecture with
   three distinct roles: Agent Root, Agent Registry and Agent Resolver.
   This architecture aligns with the DNS hierarchical paradigm widely
   accepted by the internet industry,

   Each Agent holds an original FQDN as the registration prefix.  After
   registering to an Agent Registry with a fixed suffix domain, a
   globally unique composite FQDN in the form `<prefix>.`<suffix>` is
   generated, where the agents native domain becomes a subdomain of the
   registry domain.  Agent Root maintains a trusted whitelist of all
   Agent Registry suffix domains and distributes the list to all Agent
   Resolvers.  Agent Resolvers synchronize all agent metadata from
   registries via zone transfer or incremental DNS notifications, then
   build local capability database.  When an agent client requests
   agents with specific capabilities, the resolver performs capability
   matching and returns complete standardized Capability Card to the
   requester.

Yao, et al.              Expires 7 January 2027                 [Page 2]
Internet-Draft               Agent Discovery                   July 2026

   Capability Card adopts fixed JSON format to store DNS service
   records, TLS CA certificate information and capability strings.  Each
   capability description string is limited to 200 characters maximum, a
   nd multiple independent capabilities are stored as separate array
   entries.  Agent clients support two complementary mechanisms to
   discover the local Agent Resolver endpoint: static preconfiguration
   and DNS-SD dynamic lookup via reserved service label `_agent-
   index._tcp`.

2.  Terminology

   This document makes use of the following terms:

   Agent Root:  Global top-tier trusted node responsible for managing
      the whitelist of all valid Agent Registry suffix domains.

   Agent Registry:  Domain-level registration service with a fixed
      permanent suffix domain, accepting agent registration and hosting
      agent DNS records and Capability Card endpoints.

   Agent Resolver:  Distributed discovery node that stores global agent
      metadata and processes capability search requests from agent
      clients.

   Registration Prefix:  Native FQDN owned independently by an agent,
      unchanged during registration, e.g., `agent-hotel.example`.

   Registration Suffix:  Fixed domain assigned to an Agent Registry,
      e.g., `example-registry1.com`.

   Composite Agent FQDN:  Globally unique identity concatenated as
      <prefix>.<suffix>, the agent’s native domain acts as a subdomain
      of the registry domain.

   Agent Client:  Any agent instance initiating capability discovery
      requests (e.g., Agent A).

   Capability Card:  Standard JSON document storing agent DNS service
      records, TLS CA information and human-readable capability strings,
      hosted via HTTPS.

   The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
   "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and
   "OPTIONAL" in this document are to be interpreted as described in BCP
   14 [RFC2119] [RFC8174] when, and only when, they appear in all
   capitals, as shown here.

Yao, et al.              Expires 7 January 2027                 [Page 3]
Internet-Draft               Agent Discovery                   July 2026

3.  Problem Statement

   The anticipated massive deployment of Internet of Agents (IoA) will
   bring millions to billions of intelligent agents globally in the
   foreseeable future.  Efficient, scalable, and standardized discovery
   of massive distributed agents becomes a critical unresolved
   challenge.  The proposed discovery mechanisms should support fast
   lookup, unified identity management, and large-scale horizontal
   expansion for massive agent clusters.  This document specifys a DNS-
   like three-tier discovery architecture composed of three core roles:
   Agent Root, Agent Registry, and Agent Resolver.

4.  Agent Discovery Archtecture

   The agent discovery architechture is shown below:

                     Global Top Trust Layer: Agent Root
                +------------------------------------------+
                | Agent Root (Global Trust Whitelist Node) |
                | - Maintain all trusted Registry Suffixes |
                | - Periodically push registry catalog     |
                +--------------------+---------------------+
                                    ^
                                    |
                                    v  Periodic  Sync (HTTPS/DNS Notify)
+-----------------------------------------------------------------------------+
|   Middle Distributed Layer: Multiple Independent Agent Registry Nodes        |
|                                                                              |
|  +------------------------+        +------------------------+                |
|  | Agent Registry A       |        | Agent Registry B       |                |
|  | Suffix: reg1.example   |        | Suffix: reg2.example   |                |
|  | - Accept agent register|        | - Accept agent register|                |
|  | - Generate composite   |        | - Generate composite   |                |
|  |   agent FQDN           |        |   agent FQDN           |                |
|  | - Host Capability Card |        | - Host Capability Card |                |
|  +-----------+------------+        +-----------+------------+                |
|              ^                                      ^                        |
|              |                                      |                       |
|              | Zone Transfer / DNS Incremental Sync |                      |
+-----------------------------------------------------------------------------+
               |                                      |
               v                                      v
+-----------------------------------------------------------------------------+
| Bottom Discovery Layer: Distributed Agent Resolver Cluster                  |
|                                                                             |
|  +------------------------+        +------------------------+                |
|  | Agent Resolver X       |        | Agent Resolver Y       |                |

Yao, et al.              Expires 7 January 2027                 [Page 4]
Internet-Draft               Agent Discovery                   July 2026

|  | - Store all agent data |        | - Store all agent data |                |
|  | - Capability string    |        | - Capability string    |                |
|  |   matching engine      |        |   matching engine      |                |
|  | - Serve query API      |        | - Serve query API      |                |
|  +-----------+------------+        +-----------+------------+                |
|              ^                                     ^
|              |                                     |                        |
|              | Client Capability Query (HTTPS API) |                        |
+-----------------------------------------------------------------------------+
               |                                     |
               v                                     v
+---------------------------------+        +---------------------------------+
| End Client Layer: Agent Client  |        | End Client Layer: Agent Client  |
| (e.g. Agent A Query Party)      |        | (e.g. Agent A Query Party)      |
| Discovery Modes:                |        | Discovery Modes:                |
| 1. Static config resolver URL  |         | 1. Static config resolver URL  |
| 2. DNS-SD _agent-index._tcp    |         | 2. DNS-SD _agent-index._tcp    |
+---------------------------------+        +---------------------------------+

             Fig 1. Agent Discovery Archtecture

   The Fig1 shows a top-down hierarchical structure: Agent Root (top
   layer), Distributed Agent Registry Cluster (middle layer),
   Distributed Agent Resolver Cluster (bottom layer).  The end agent
   clients send capability queries to nearby local Resolvers.

   Agent Root is the unique global trusted root node with mandatory
   responsibilities: 1) Receive enrollment applications from new Agent
   Registries and complete domain ownership validation; 2) Maintain a
   persistent trusted whitelist of all registry suffix domains; 3)
   Periodically push the full whitelist to online Agent Resolvers; 4)
   Broadcast revocation signals to all Resolvers when a registry is
   compromised or malicious.

   Mandatory storage constraint: Agent Root MUST NOT store any business
   agent data, only lightweight registry metadata including suffix
   domain, operator information and sync endpoint to guarantee high
   availability.

Yao, et al.              Expires 7 January 2027                 [Page 5]
Internet-Draft               Agent Discovery                   July 2026

   Agent Registry owns a fixed permanent suffix domain.  The core duties
   are : 1) Process agent registration, update and unregistration
   requests; 2) Generate composite global unique FQDN by concatenating
   agent prefix and registry suffix; 3) Publish standard A, SVCB and SRV
   records on local authoritative DNS servers for registered agents; 4)
   Expose public HTTPS endpoints to serve each agent’s Capability Card;
   5) Provide zone transfer and DNS incremental sync interfaces for
   downstream Agent Resolvers; 6) Maintain agent registration expiration
   time and online status.

   Agent Resolver acts as the unified public entry for all capability
   queries.  The core duties are : 1) Periodically pull the latest
   trusted registry whitelist from Agent Root; 2) Establish persistent
   synchronization connections with each valid registry to pull all
   agent metadata; 3) Build local index mapping composite agent FQDN to
   full Capability Card and capability string list; 4) Accept capability
   search requests from agent clients and execute capability matching;
   5) Return complete Capability Card records of all matched agents to
   query clients.

5.  Capability Card Format

   Capability Card is a mandatory fixed-format JSON file hosted via
   HTTPS for every registered agent.  Format constraints:

Yao, et al.              Expires 7 January 2027                 [Page 6]
Internet-Draft               Agent Discovery                   July 2026

- Each single capability string inside `capability_list`
  MUST NOT exceed 200 UTF-8 characters;
- Multiple independent service capabilities are stored
  as separate array items;
- Integrates SVCB/SRV service records, TLS CA fingerprint
   and certificate validity window.

Practical example:
    Prefix = agent-hotel.example
    Suffix = example-registry1.com
    Composite global unique agent identity: `agent-hotel.example.example-registry1.com

Template:
json
{
  "agent_fqdn": "agent-hotel.example.example-registry1.com",
  "dns_service_info": {
    "svcb_record": "SVCB 0 mcp agent-hotel.example.example-registry1.com port=443 alpn=h2",
    "srv_record": "_mcp._tcp.agent-hotel.example.example-registry1.com"
  },
  "network_endpoint": "https://agent-hotel.example.example-registry1.com/mcp/v1",
  "tls_ca_info": {
    "ca_sha256_fingerprint": "E3:4F:2A:89:11:CC:90:7B:...",
    "cert_valid_until": "2027-12-31T23:59:59Z"
  },
  "capability_list": [
    "capability1: Provide hotel room online booking, and order status query service",
    "capability2: Provide real-time hotel room price inquiry service",
    "capability3: Support hotel order cancellation and after-sales processing service"
  ],
  "agent_online_status": "online",
  "register_expire_time": "2027-12-31T23:59Z"
}

6.  DNS-like Design

   The term "DNS-like" used throughout this document indicates that the
   proposed three-tier agent discovery architecture borrows core design
   paradigms from the Domain Name System (DNS).  Specifically, the
   architecture adopts three key DNS design concepts: 1) DNS
   hierarchical layered partitioning: It follows DNS’s tree-based tiered
   delegation model to split global agent identity management into root,
   registry, and resolver layers, enabling distributed horizontal
   scaling of agent metadata. 2)DNS-style distributed data management:
   Similar to how DNS separates authoritative data storage across
   disjoint zones, agent registration information is decentralised and

Yao, et al.              Expires 7 January 2027                 [Page 7]
Internet-Draft               Agent Discovery                   July 2026

   maintained by dedicated registry nodes without a single centralised
   bottleneck. 3) DNS-inspired asynchronous data replication: The
   solution employs zone-data synchronisation mechanisms analogous to
   DNS zone transfers, allowing incremental, low-overhead propagation of
   agent registration records across hierarchical nodes.

   Communication between the Agent Root, Agent Registry, and Agent
   Resolver components of this DNS-like three-tier agent discovery
   architecture employs standard DNS protocol for data exchange.
   Metadata associated with agents is stored and carried in the form of
   DNS resource records across all hierarchical nodes.  This design
   choice leverages DNS’s well-specified wire format, mature message
   processing logic, and proven zone s ynchronization mechanisms to
   facilitate distributed agent information lookup and cross-layer data
   replication.

7.  Congigure Agent Client to Discover Local Resolver

   Agent clients such as Agent A need to acquire the HTTPS endpoint
   address of the local Agent Resolver before sending capability search
   requests.  Two discovery mechanisms are defined, with DNS-SD as the
   recommended primary mode and static preconfiguration as fallback.

8.  IANA Considerations

   to be added.

9.  Security Considerations

   to be added

10.  Acknowledgements

   to be added

11.  Normative References

   [RFC2119]  Bradner, S., "Key words for use in RFCs to Indicate
              Requirement Levels", BCP 14, RFC 2119,
              DOI 10.17487/RFC2119, March 1997,
              <https://www.rfc-editor.org/info/rfc2119>.

   [RFC1034]  Mockapetris, P., "Domain names - concepts and facilities",
              STD 13, RFC 1034, DOI 10.17487/RFC1034, November 1987,
              <https://www.rfc-editor.org/info/rfc1034>.

Yao, et al.              Expires 7 January 2027                 [Page 8]
Internet-Draft               Agent Discovery                   July 2026

   [RFC1035]  Mockapetris, P., "Domain names - implementation and
              specification", STD 13, RFC 1035, DOI 10.17487/RFC1035,
              November 1987, <https://www.rfc-editor.org/info/rfc1035>.

   [RFC9665]  Lemon, T. and S. Cheshire, "Service Registration Protocol
              for DNS-Based Service Discovery", RFC 9665,
              DOI 10.17487/RFC9665, June 2025,
              <https://www.rfc-editor.org/info/rfc9665>.

   [RFC8174]  Leiba, B., "Ambiguity of Uppercase vs Lowercase in RFC
              2119 Key Words", BCP 14, RFC 8174, DOI 10.17487/RFC8174,
              May 2017, <https://www.rfc-editor.org/info/rfc8174>.

Appendix A.  Change History

A.1.  version 00

   1.  Initial draft to propose the Agent Discovery Architecture.

Authors' Addresses

   Jiankang Yao
   CNNIC
   4 South 4th Street,Zhongguancun,Haidian District
   Beijing
   Beijing, 100190
   China
   Phone: +86 10 59116505
   Email: yaojk@cnnic.cn

   Guanggang
   Jinan University
   China
   Email: gggeng@jnu.edu.cn

   Meiling
   China Mobile
   China
   Email: chenmeiling@chinamobile.com

Yao, et al.              Expires 7 January 2027                 [Page 9]
Internet-Draft               Agent Discovery                   July 2026

   Hongtao Li
   CNNIC
   4 South 4th Street,Zhongguancun,Haidian District
   Beijing
   Beijing, 100190
   China
   Email: lihongtao@cnnic.cn

Yao, et al.              Expires 7 January 2027                [Page 10]