Network Working Group L. Jakab
Internet-Draft F. Brockners
Intended status: Informational Cisco Systems
Expires: 1 May 2027 28 October 2026
Zero-Configuration Agent Discovery
draft-jakab-dawn-agent-discovery-mdns-01
Abstract
Protocols for communication between autonomous software agents
typically discover agents through documents retrieved over HTTPS from
a well-known URI. This model requires a discovering party to already
know an agent's host name or URL, or to consult a centralized
catalog. It provides no zero-configuration mechanism for enumerating
agents that are present on a local network.
This document describes how existing, widely deployed protocols,
Multicast DNS (mDNS) and DNS-Based Service Discovery (DNS-SD), can be
used to advertise and discover agents on a local link, with no new
protocol machinery. It presents a general, protocol-independent two-
stage discovery model in which lightweight enumeration over DNS-SD is
followed by obtaining full agent metadata over the agent's native
transport, whether by retrieving a description document or by
connecting to the agent and performing its native handshake. The
DNS-SD records carry a small set of discovery parameters that is
shared with agent discovery in unicast DNS. The document gives
worked instantiations of that model for the Agent2Agent (A2A)
protocol and the Model Context Protocol (MCP).
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 1 May 2027.
Jakab & Brockners Expires 1 May 2027 [Page 1]
Internet-Draft Zero-Configuration Agent Discovery October 2026
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
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 . . . . . . . . . . . . . . . . . . . . . . . . 3
2. Terminology . . . . . . . . . . . . . . . . . . . . . . . . . 4
3. Use Cases for Discovery of Local Agents . . . . . . . . . . . 5
4. Requirements for Discovery of Local Agents . . . . . . . . . 6
5. DNS-SD and mDNS-Based Solution Approach . . . . . . . . . . . 8
5.1. Agent or Agent Resource Discovery over DNS-SD . . . . . . 9
5.1.1. Carrying the Descriptor: TXT and SVCB . . . . . . . . 11
5.2. Two-Stage Discovery . . . . . . . . . . . . . . . . . . . 11
5.2.1. Stage 1: Browse (DNS-SD) . . . . . . . . . . . . . . 11
5.2.2. Stage 2: Obtain Description and Connect . . . . . . . 12
6. Instantiations . . . . . . . . . . . . . . . . . . . . . . . 12
6.1. A2A . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
6.1.1. Service Type and Protocol Identifier . . . . . . . . 13
6.1.2. AgentCard-to-DNS-SD Mapping . . . . . . . . . . . . . 13
6.1.3. Service Instance Name Encoding and Conflicts . . . . 15
6.1.4. TXT Record Keys . . . . . . . . . . . . . . . . . . . 16
6.1.5. Example . . . . . . . . . . . . . . . . . . . . . . . 18
6.2. MCP . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
6.2.1. Service Type and Protocol Identifier . . . . . . . . 20
6.2.2. MCP-to-DNS-SD Mapping . . . . . . . . . . . . . . . . 20
6.2.3. Service Instance Name Encoding and Conflicts . . . . 22
6.2.4. TXT Record Keys . . . . . . . . . . . . . . . . . . . 23
6.2.5. Example . . . . . . . . . . . . . . . . . . . . . . . 24
6.3. Other Instantiations . . . . . . . . . . . . . . . . . . 24
7. Deployment Considerations . . . . . . . . . . . . . . . . . . 25
7.1. Discovery Beyond the Local Link . . . . . . . . . . . . . 25
8. Related work . . . . . . . . . . . . . . . . . . . . . . . . 26
8.1. LAD-A2A project . . . . . . . . . . . . . . . . . . . . . 26
8.2. MCP DNS discovery . . . . . . . . . . . . . . . . . . . . 26
8.3. OpenClaw project . . . . . . . . . . . . . . . . . . . . 27
9. Security Considerations . . . . . . . . . . . . . . . . . . . 27
10. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 29
Jakab & Brockners Expires 1 May 2027 [Page 2]
Internet-Draft Zero-Configuration Agent Discovery October 2026
11. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . 30
12. References . . . . . . . . . . . . . . . . . . . . . . . . . 30
12.1. Normative References . . . . . . . . . . . . . . . . . . 30
12.2. Informative References . . . . . . . . . . . . . . . . . 30
Appendix A. Changes Since Previous Revisions . . . . . . . . . . 32
A.1. Since -00 . . . . . . . . . . . . . . . . . . . . . . . . 32
Contributors . . . . . . . . . . . . . . . . . . . . . . . . . . 33
Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 34
1. Introduction
AI agents are increasingly being deployed not only as cloud-hosted
services, but also as local services running within constrained
environments such as developer workstations, laboratories, branch
offices, industrial sites, meeting rooms, smart buildings, edge
deployments, and other operationally controlled networks. In these
environments, agents can provide capabilities such as local
inference, device control, data access, workflow automation,
observability, robotics coordination, or interaction with nearby
tools and systems.
For such deployments, agents, clients, orchestrators, and tools need
a way to discover one another without relying on manual
configuration, fixed addressing, or continuous connectivity to a
central registry. Agents can be started and stopped dynamically,
moved between hosts, assigned different network addresses, or
deployed temporarily for a specific task, experiment, site, or
operational context. A discovery mechanism can reduce configuration
burden, improve resilience, and allow local agent ecosystems to form
and adapt within the boundaries of the environment in which they
operate.
Agent discovery in these environments has a specific set of
requirements. The discovery mechanism needs to be local in scope,
suitable for networks ranging from a single link to a site, and
usable in networks that are operationally controlled and at least
somewhat trusted. It needs to support zero- or low-configuration
operation, allow agents to advertise basic service availability, and
enable clients to obtain enough information to connect to a more
complete agent description or capability manifest. At the same time,
discovery ought to avoid exposing sensitive details unnecessarily,
ought not be treated as a substitute for authentication or
authorization, and ought to allow deployments to apply local policy,
segmentation, and access control.
Agent communication protocols, including the Agent2Agent (A2A)
protocol [A2A] or the Model Context Protocol (MCP) [MCP], generally
assume that the party that wishes to connect to an agent or agent
Jakab & Brockners Expires 1 May 2027 [Page 3]
Internet-Draft Zero-Configuration Agent Discovery October 2026
resource already possesses the resource's host name or URL, or can
query a catalog or registry service. Such protocols do not generally
provide a mechanism for a party to enumerate agents that are
reachable on the local network without prior knowledge of their
addresses. This is a common situation when a person, device, or
agent enters an unfamiliar local environment, such as a building,
venue, home, lab, or operational site, whose local services may be
fronted by agents.
Because these environments are local, constrained, and operationally
managed, a DNS-Based Service Discovery approach may be attractive.
In particular, DNS-SD, and where appropriate its use with mDNS,
provides a well-established model for discovering services on local
networks without requiring prior knowledge of host addresses or a
central discovery service. This document explores how a DNS-SD based
mechanism could be applied to local agent discovery, while preserving
the separation between discovery, agent description, and trust
establishment.
This document is Informational. It does not modify, and does not
define a new profile of, the A2A protocol, the MCP protocol, the DNS
protocol, mDNS, or DNS-SD. It describes one possible use of these
existing protocols for agent discovery.
2. Terminology
This document uses the DNS-SD terminology of [RFC6763], in particular
"service type", "service instance", "service instance name",
"browse", and "resolve". It uses "AgentCard", "agent", and "session"
as described in [A2A], and "MCP server", "MCP endpoint", and
"initialize" as described in [MCP]. Where this document refers
generically to an advertised entity, the term "agent" is used to
include an MCP server.
This document adopts the agent-discovery terminology of
[I-D.farrel-dawn-terminology] where it applies. In particular, the
minimal descriptor that this document carries in DNS-SD records,
defined in Section 5.1, is an instance of the "Minimum Discoverable
Information" (MDI) of [I-D.farrel-dawn-terminology]: the minimum
amount of information an entity needs to provide to become
discoverable, acting as the common header that fronts the agent's
fuller description. In the two-stage flow of Section 5.2, this MDI
is the first-stage information that locates a local agent and points
to the full description obtained in the second stage.
For clarity, this document uses the following roles:
advertising agent: an agent that publishes a DNS-SD service instance
Jakab & Brockners Expires 1 May 2027 [Page 4]
Internet-Draft Zero-Configuration Agent Discovery October 2026
for itself, directly or through an existing DNS-SD responder.
discovering party: a client, often itself an agent acting on behalf
of a user, that browses for an agent service type to enumerate
advertising agents.
3. Use Cases for Discovery of Local Agents
Local agent discovery can be useful in a range of use cases. In each
of them, a client does not know in advance which agents are present,
and the set changes with location and over time, so it cannot be
configured ahead of time. A common theme across these use cases is
that they involve operationally controlled and at least somewhat
trusted networks, ranging from a single link to a site.
* *Visitors in shops, venues, and offices:* A personal agent on a
phone or laptop that joins the network of a shop, venue, or office
may need to discover the agents that serve that place, such as
wayfinding, ordering, or meeting services. The visitor knows
neither their names nor any directory in advance.
* *Developer and test environments:* A developer workstation, lab
network, or test bench may run multiple local agents, such as
coding agents, test agents, local model servers, browser
automation agents, or tool adapters. Discovery reduces the need
for static configuration and allows agents to be started, stopped,
or moved dynamically.
* *Local tool and capability servers:* An agent runtime may need to
discover nearby tool servers, such as file indexers, database
proxies, source-control helpers, observability tools, or device-
control interfaces. Discovery enables an agent to locate
available tools and then retrieve a more complete manifest or
capability description.
* *Edge and branch deployments:* A branch office, warehouse, retail
site, factory cell, or other edge location may contain agents for
local inference, sensor processing, caching, monitoring, or
automation. Local discovery allows agents and orchestrators to
operate even when connectivity to a central registry or cloud
service is unavailable.
* *Meeting-room and collaboration systems:* A meeting assistant or
collaboration agent may need to discover room-local services, such
as display agents, transcription agents, camera-control agents,
whiteboard agents, or room-control systems. Discovery supports
low-configuration pairing within a physical space.
Jakab & Brockners Expires 1 May 2027 [Page 5]
Internet-Draft Zero-Configuration Agent Discovery October 2026
* *Smart home and smart building environments:* A local automation
agent may need to discover agents associated with lighting,
climate, energy management, access control, appliances, or
building systems. Discovery is useful because devices may join,
leave, or change addresses over time.
* *Robotics and autonomous systems labs:* A planner, controller, or
operator agent may need to discover robot capability agents, such
as navigation, mapping, perception, manipulation, docking, or
telemetry services. Discovery helps support mobile or modular
systems that join and leave the local network.
* *Industrial and operational technology environments:* A site-local
agent may need to discover equipment-facing agents, telemetry
agents, digital-twin interfaces, or inspection systems within an
industrial cell or plant-floor segment.
* *Temporary, offline, or incident-response networks:* Field
deployments, incident-response kits, emergency operations, or
temporary collaboration networks may lack reliable DNS, Internet
connectivity, or central registries. Local discovery can allow
agents to find one another with minimal infrastructure.
* *Vehicle, drone, and onboard networks:* An onboard orchestrator
may need to discover local agents providing perception,
diagnostics, planning, sensor access, or compute services.
Discovery can support modular systems and component replacement
without fixed addressing.
* *Local inventory and observability:* An operator or management
agent may need to determine which agent services are present on a
local network segment. Discovery can provide a lightweight view
of locally available agent endpoints and their basic service
types.
For a scenario that combines local discovery, as described in this
document, with global discovery in unicast DNS, see the stadium use
case in [I-D.blwm-dawn-common-discovery-parameters].
4. Requirements for Discovery of Local Agents
This section distills discovery-specific requirements derived from
the use cases above. The requirements apply to the mechanism by
which a client, orchestrator, tool, or other agent learns that a
local agent service exists and obtains enough information to initiate
further interaction.
Jakab & Brockners Expires 1 May 2027 [Page 6]
Internet-Draft Zero-Configuration Agent Discovery October 2026
Requirements related to authentication, authorization, task
execution, safety policy, or agent behavior after discovery are
outside the scope of discovery.
The requirements below are stated descriptively; this document is
Informational and uses no normative (BCP 14) language.
* *REQ-1: Local discovery scope:* The discovery mechanism needs to
support discovery of agent services within a local network or
other bounded operational domain. It should not require Internet
reachability.
* *REQ-2: Low-configuration operation:* The discovery mechanism
should minimize the need for static configuration, fixed IP
addresses, preconfigured host names, or manually maintained
endpoint lists.
* *REQ-3: Operation without central infrastructure:* The discovery
mechanism should be able to operate when a central registry, cloud
service, or site-wide directory is unavailable. This is
particularly important for temporary, offline, edge, vehicle, and
incident-response networks.
* *REQ-4: Dynamic availability:* The discovery mechanism needs to
support agents that appear, disappear, restart, move between
hosts, or change network addresses. The mechanism needs to
provide a way for clients to determine that previously discovered
information is no longer valid.
* *REQ-5: Multiple agents and multiple instances:* The discovery
mechanism needs to support multiple agent services on the same
host, multiple hosts on the same local network, and multiple
instances of the same agent service type.
* *REQ-6: Basic rendezvous information:* The discovery mechanism
needs to provide enough information for a client to initiate
contact with a discovered agent service. This includes, or
enables resolution of, the service endpoint, transport, and port
or equivalent connection information.
* *REQ-7: Service type identification:* The discovery mechanism
should allow clients to distinguish different classes of agent
services, such as agent runtimes, tool servers, model-serving
agents, device-control agents, observability agents, or capability
catalogs.
Jakab & Brockners Expires 1 May 2027 [Page 7]
Internet-Draft Zero-Configuration Agent Discovery October 2026
* *REQ-8: Minimal discovery metadata:* The discovery mechanism
should support limited metadata sufficient for selection and
bootstrapping, such as a protocol identifier, version, basic
locality hint, or pointer to a more complete agent description.
Discovery advertisements should avoid exposing detailed
capabilities, sensitive operational context, user data, topology
information, or security-sensitive configuration.
* *REQ-9: Rendezvous to richer agent description:* The discovery
mechanism should allow a discovered service to provide enough
information to reach a more complete agent description or
capability set, whether that description is a retrievable manifest
or document, or is obtained by connecting to the service and
performing its native handshake. Retrieval or connection, format,
access control, and semantics of that description are separate
from basic discovery unless specified elsewhere.
* *REQ-10: Local policy compatibility:* The discovery mechanism
should be compatible with local administrative controls, including
network segmentation, service filtering, interface selection, and
restrictions on which services are advertised or discoverable.
* *REQ-11: Support for constrained operational environments:* The
discovery mechanism should be usable in environments with limited
administrative overhead, intermittent connectivity, changing
device membership, and local networks ranging from a single link
to a site.
* *REQ-12: Discovery freshness:* The discovery mechanism needs to
provide a way for clients to reason about the freshness of
discovered information, for example through lifetimes, expiry,
withdrawal, or rediscovery.
A more general set of requirements for discovery of agents and agent
related resources is found in [I-D.king-dawn-requirements].
5. DNS-SD and mDNS-Based Solution Approach
DNS-Based Service Discovery (DNS-SD) [RFC6763] over Multicast DNS
(mDNS) [RFC6762] can address the discovery requirements listed above.
DNS-SD over mDNS is a common zero-configuration service discovery
technology on local links. It allows a service to advertise its
presence, and a client to enumerate available service instances, with
no pre-configuration and no central server. DNS-SD is, however, a
general framework: by itself it neither names a service type for
agents nor describes how to represent agent metadata in DNS-SD
records.
Jakab & Brockners Expires 1 May 2027 [Page 8]
Internet-Draft Zero-Configuration Agent Discovery October 2026
This document describes how DNS-SD and mDNS can be used to discover
agents. It does not define new protocol elements. It presents:
* a general, protocol-independent model for advertising and
discovering agents over DNS-SD, comprising only the "_agents._tcp"
and "_agents._udp" service types, with per-protocol subtypes, and
a minimal descriptor carried in DNS-SD records, with the agent
protocol named in a "proto" key and the descriptor built from the
common discovery parameters of
[I-D.blwm-dawn-common-discovery-parameters] (Section 5.1);
* a two-stage discovery flow that separates lightweight local
enumeration from full metadata retrieval (Section 5.2);
* worked instantiations of the model for A2A and MCP (Section 6).
The model is intended to accommodate further agent protocols.
The approach is intentionally compatible with existing DNS-SD tooling
and infrastructure. It introduces no new DNS resource record types
and requires no changes to mDNS, DNS-SD, or any agent protocol.
5.1. Agent or Agent Resource Discovery over DNS-SD
The approach described in this document rests on a small, protocol-
independent pattern. It has three elements:
1. A single, protocol-independent DNS-SD service type,
"_agents._tcp", that identifies agents on the local link
regardless of which agent protocol they speak. The specific
agent protocol is not encoded in the service type; instead it is
carried in a "proto" TXT key (see element 2), so that a
discovering party browses once for "_agents._tcp" and learns the
protocol of each discovered agent from its descriptor. A
discovering party interested in only one agent protocol can
instead browse a per-protocol subtype ([RFC6763], Section 7.1),
such as "_a2a._sub._agents._tcp" or "_mcp._sub._agents._tcp"; a
subtype names the same protocol as the "proto" key of the
instances it lists. Following [RFC6763], Section 7, the service
name is "agents", and the second label of the service type names
the transport. Agents reachable over TCP use "_agents._tcp";
this includes HTTP-based agents that also offer HTTP/3 over QUIC,
which list "h3" in the "alpn" key. The "_agents._udp" service
type accommodates agents that are reachable only over a transport
other than TCP, such as an agent protocol that runs solely over
QUIC; no such protocol is in use today, so this document uses
"_agents._tcp" throughout.
Jakab & Brockners Expires 1 May 2027 [Page 9]
Internet-Draft Zero-Configuration Agent Discovery October 2026
2. A minimal descriptor for each advertised agent, carried in the
DNS-SD service instance: the instance name, an SRV record giving
the host and port, and exactly one TXT record carrying a small
set of key/value pairs. The keys are the common discovery
parameters defined in
[I-D.blwm-dawn-common-discovery-parameters], which uses the same
parameters for DNS-SD and for unicast DNS. The descriptor always
begins with a "txtvers" key ([RFC6763], Section 6.7) and includes
a "proto" key naming the agent protocol (for example,
"proto=a2a"), together with enough rendezvous information to
reach the agent's full, authoritative description over the agent
protocol's own transport. It is deliberately kept small, and is
the Minimum Discoverable Information (MDI) for an agent on the
local link: it carries only what a discovering party needs to
enumerate candidate agents, learn their protocol, and perform
lightweight selection. The parameters used for each protocol are
described in Section 6.
3. A two-stage discovery flow (Section 5.2): the DNS-SD descriptor
is used for local enumeration, and the full agent description is
then obtained over the agent protocol's own transport. Depending
on the protocol, this is done either by retrieving a description
document (as in A2A) or by connecting to the agent and performing
its native handshake (as in MCP).
An agent that offers more than one agent protocol is advertised as
one service instance per protocol, each with its own SRV and TXT
records, rather than as a single instance listing several protocols.
When the instances carry the same "desc-ref" value, a reference to an
external descriptor of the agent, a discovering party that finds
several of them can infer that they belong to the same agent.
[I-D.blwm-dawn-common-discovery-parameters] gives an example of such
an agent.
The value of the pattern is that it reuses zero-configuration
discovery and keeps agent-specific descriptions in the agent's native
description format rather than attempting to encode them in DNS. A
single "_agents._tcp" service type lets a discovering party enumerate
all local agents in one browse operation, independent of protocol,
and then select among them using the "proto" key and other hints.
Section 6 instantiates the pattern for A2A and for the Model Context
Protocol as examples; the same pattern applies to further agent
protocols by assigning a "proto" value and the corresponding subtype
name.
Jakab & Brockners Expires 1 May 2027 [Page 10]
Internet-Draft Zero-Configuration Agent Discovery October 2026
5.1.1. Carrying the Descriptor: TXT and SVCB
This document uses a TXT record ([RFC6763], Section 6) as the default
carrier for the descriptor's key/value pairs, because TXT records are
supported by essentially all deployed mDNS/DNS-SD implementations,
which is what makes the approach usable with no new software.
Service Binding (SVCB) records [RFC9460] can carry the same
connection and parameter information in a more structured form;
[I-D.gakiwate-dnssd-use-svcb] describes how DNS-SD clients can use
SVCB records to learn endpoint properties, such as ALPN identifiers.
A deployment in which all participating implementations are known to
support SVCB over mDNS may use SVCB for greater efficiency. Because
SVCB support in mDNS responders and clients is not yet widespread,
this document keeps TXT as the interoperable default and treats SVCB
as a forward-looking option. This is the reverse of the wide-area
DNS-AID approach [I-D.mozleywilliams-dnsop-dnsaid], where SVCB is the
primary carrier and TXT a fallback; supporting SVCB here aids
compatibility with such wide-area mechanisms. The key names are the
same in both carriers: [I-D.blwm-dawn-common-discovery-parameters]
defines each parameter for both TXT and SVCB, so a discovering party
needs only one vocabulary, and documents the considerations for SVCB
in detail, such as the SvcParamKey codes that remain private-use
until IANA assigns them. Section 6.1.5 shows an A2A agent advertised
with TXT and, for illustration, with SVCB.
5.2. Two-Stage Discovery
Discovery proceeds in two stages. The first uses DNS-SD for
lightweight, zero-configuration enumeration on the local link. The
second uses the agent's native transport to retrieve the complete
agent description and to begin a session. This follows the common
DNS-SD pattern in which discovery provides rendezvous information and
application protocols perform subsequent interaction.
5.2.1. Stage 1: Browse (DNS-SD)
A discovering party browses for "_agents._tcp.local.", or for a per-
protocol subtype such as "_a2a._sub._agents._tcp.local.", and, for
each advertised instance, resolves the SRV and TXT records. From the
descriptor it learns the agent protocol (the "proto" key), a
reference to an external descriptor of the agent, if any (the "desc-
ref" key), the rendezvous information for reaching the full
description, and any lightweight selection hints. This is enough to
filter candidate agents, including by protocol, without obtaining
metadata over the agent's transport. Selecting agents by what they
can do, such as their skills or tools, is left to the second stage,
where the agent protocol describes these authoritatively. Carrying
Jakab & Brockners Expires 1 May 2027 [Page 11]
Internet-Draft Zero-Configuration Agent Discovery October 2026
capabilities in DNS-SD records would need a shared vocabulary, would
change the records whenever the capabilities change, and would expose
details that advertisements should avoid (REQ-8); see
[I-D.blwm-dawn-common-discovery-parameters], Appendix B.3. If coarse
categories prove necessary, DNS-SD subtypes can provide them, as the
per-protocol subtypes do today.
5.2.2. Stage 2: Obtain Description and Connect
For a selected candidate, the discovering party obtains the full
agent description over the agent protocol's own transport, using the
rendezvous information from the descriptor. The form this takes is
protocol-specific: for A2A, the party retrieves and validates the
AgentCard from a description document; for MCP, it connects to the
MCP endpoint and performs the "initialize" handshake. It then
selects the protocol-specific connection details (for A2A, an entry
from supportedInterfaces), selects an authentication method, and
initiates a session according to the agent protocol [A2A] [MCP].
From this point, interaction is governed entirely by the agent
protocol; DNS-SD plays no further role.
Where the agent protocol has a description document, the descriptor
can carry a "desc-ref-digest" key, a digest of that document. A
discovering party that already holds a description whose digest
matches can reuse it instead of retrieving it again; a mismatch means
only that the description has changed.
6. Instantiations
This section instantiates the model of Section 5.1 for specific agent
protocols. Protocols for communication between autonomous software
agents are emerging rapidly. This document gives two instantiations
as examples: the Agent2Agent (A2A) protocol [A2A] (Section 6.1) and
the Model Context Protocol (MCP) [MCP] (Section 6.2).
The two differ in an instructive way. In A2A, an agent publishes a
description of itself, an AgentCard, as a JSON document served over
HTTPS at a well-known URI, for example, "/.well-known/agent-
card.json"; a party wishing to interact with an agent retrieves the
AgentCard, selects an authentication method, and initiates a session.
In MCP, there is no equivalent static description document: a party
connects to the server's endpoint and learns the server's identity,
protocol version, and capabilities from the live protocol handshake.
The model accommodates both, because the second stage (Section 5.2)
is defined as retrieval of the authoritative description by the agent
protocol's own means, whether that is fetching a document or
performing a handshake.
Jakab & Brockners Expires 1 May 2027 [Page 12]
Internet-Draft Zero-Configuration Agent Discovery October 2026
6.1. A2A
This section describes the model of Section 5.1 for version 1.0.1 of
the A2A protocol [A2A]. In that version, an AgentCard declares
endpoint interfaces through the ordered supportedInterfaces list.
This section is descriptive and does not define a required A2A
encoding.
6.1.1. Service Type and Protocol Identifier
An A2A agent that participates in local discovery is advertised as a
DNS-SD service instance of the protocol-independent service type
"_agents._tcp" (Section 5.1), in the "local." domain when using mDNS
[RFC6762], and of its "_a2a" subtype. It is identified as an A2A
agent by the descriptor key "proto=a2a" in its TXT record (see
Section 6.1.4). No A2A-specific service type is defined; a
discovering party distinguishes A2A agents from agents of other
protocols by the "proto" key, or enumerates only A2A agents by
browsing "_a2a._sub._agents._tcp".
6.1.2. AgentCard-to-DNS-SD Mapping
The descriptor projects a minimal subset of AgentCard fields and
associated publication information onto a DNS-SD service instance.
It carries only what is useful for enumeration and lightweight
selection; everything else is obtained by retrieving the full
AgentCard (see Section 5.2). The mapping described in this document
is:
Jakab & Brockners Expires 1 May 2027 [Page 13]
Internet-Draft Zero-Configuration Agent Discovery October 2026
+=================+=================+===============================+
| Source field or | DNS-SD record | Transformation |
| value | | |
+=================+=================+===============================+
| (protocol | TXT key proto, | Fixed value a2a, |
| identifier) | subtype _a2a | marking the instance as |
| | | an A2A agent |
+-----------------+-----------------+-------------------------------+
| AgentCard name | Service | Encoded as the instance |
| | instance name | label (see |
| | | Section 6.1.3) |
+-----------------+-----------------+-------------------------------+
| AgentCard name | TXT key title | Copied when the |
| | | instance name differs |
| | | from it |
+-----------------+-----------------+-------------------------------+
| AgentCard URL | SRV target and | Host name and port at |
| authority | port | which the AgentCard is |
| | | served |
+-----------------+-----------------+-------------------------------+
| AgentCard URL | TXT key well- | Suffix after "/.well- |
| path | known | known/"; omitted when |
| | | the AgentCard is at the |
| | | A2A well-known path |
+-----------------+-----------------+-------------------------------+
| AgentCard | TXT key desc- | Algorithm-prefixed |
| document | ref-digest | digest of the |
| | | AgentCard, as a |
| | | freshness signal |
+-----------------+-----------------+-------------------------------+
| (external | TXT key desc- | Reference to an |
| descriptor) | ref | external descriptor of |
| | | the agent, such as an |
| | | ARD URN or an ANS name |
+-----------------+-----------------+-------------------------------+
| interface | TXT key proto- | An A2A protocol version |
| protocolVersion | ver | the agent supports, as |
| | | an optional selection |
| | | hint |
+-----------------+-----------------+-------------------------------+
| all other | (not | Obtained by retrieving |
| fields | advertised) | the AgentCard (see |
| | | below) |
+-----------------+-----------------+-------------------------------+
Table 1
Jakab & Brockners Expires 1 May 2027 [Page 14]
Internet-Draft Zero-Configuration Agent Discovery October 2026
The SRV record gives the host name and port at which the AgentCard is
served. Following [RFC6763], Section 6.3, they are not duplicated in
the TXT record. A discovering party reconstructs the AgentCard URL
from the SRV target host name and port and the well-known path, using
the "https" scheme. When the "well-known" key is absent, the path is
the A2A well-known AgentCard path "/.well-known/agent-card.json";
when it is present, the path is "/.well-known/" followed by its
value. For example, an SRV record giving host "room1.local" and port
8002, with no "well-known" key, yields the AgentCard URL
"https://room1.local:8002/.well-known/agent-card.json". If the port
is the default for the scheme, an implementation may omit it from the
reconstructed URL. An advertising implementation therefore only
needs to include "well-known" when the AgentCard is served at a non-
default location.
The A2A endpoint URLs are not advertised. They are taken from the
supportedInterfaces list of the retrieved AgentCard, which is
authoritative for A2A interaction, and they need not share the
AgentCard's host name or port. The protocol binding of each
interface, and other per-interface values such as tenant, are
likewise learned from the AgentCard. The optional "proto-ver" key is
only a lightweight selection hint: a discovering party still
retrieves the AgentCard and performs A2A interface selection using
its supportedInterfaces list.
6.1.3. Service Instance Name Encoding and Conflicts
The AgentCard name is used as the initial candidate for the
"Instance" portion of the DNS-SD service instance name
(<Instance>._agents._tcp.local.). The A2A specification places no
limits on the length or characters of name, while the instance
portion is a single DNS label of UTF-8 text in Unicode Normalization
Form C, at most 63 octets long and free of ASCII control characters
([RFC6763], Section 4.1.1); with non-ASCII characters taking up to
four octets each, as few as 15 characters may fit. The advertising
implementation therefore normalizes the name to Normalization Form C
before checking it. If the name is empty, contains ASCII control
characters, or is longer than 63 octets once encoded, the advertising
implementation does not silently truncate or alter it. Instead, it
either does not advertise the instance or uses an explicitly
configured alternate DNS-SD instance name that meets these rules.
With an alternate instance name, it can still carry an AgentCard name
that is only too long in the "title" key, which allows up to 249
octets, so that a discovering party can display it.
Literal dots and backslashes in the AgentCard name are part of the
DNS-SD instance label, not label separators. Implementations that
construct or parse DNS presentation-form names need to preserve label
Jakab & Brockners Expires 1 May 2027 [Page 15]
Internet-Draft Zero-Configuration Agent Discovery October 2026
boundaries. When the instance name is rendered in DNS presentation
form, literal dots and backslashes in the instance portion are
escaped following [RFC6763], Section 4.3. For example, the instance
portion Lab.Agent\One is represented in a full presentation-form
service instance name as Lab\.Agent\\One._agents._tcp.local..
If two agents attempt to advertise the same service instance name on
the same link, the mDNS conflict-resolution behavior of [RFC6762]
applies. An agent that registers with SRP (Section 7.1) meets the
same situation when the registrar reports that another agent already
holds the name ([RFC9665], Section 3.2.5.2). An advertising
implementation that loses the name conflict either chooses a new
unique DNS-SD instance name, probes or registers again, and
advertises using that name, or fails to advertise if local policy
forbids automatic renaming. The final service instance name can
therefore differ from the AgentCard name. The advertising
implementation can then include the "title" key with the AgentCard
name, so that a discovering party can display the agent's name
without retrieving the AgentCard. The retrieved AgentCard remains
the authoritative source for the agent's name and other A2A metadata.
A discovering party that needs to recognize the same agent across
renamed instances, or across the instances of an agent that offers
several protocols, relies on a shared "desc-ref" value, where
present, rather than on the instance name or the "title" key.
Instance names are unique only within a link; when advertisements
from several links are made visible together (Section 7.1), different
agents can appear with the same instance name, and a discovering
party can tell them apart by their "desc-ref" values, where present.
6.1.4. TXT Record Keys
The TXT record carries the key/value pairs in Table 2, encoded per
[RFC6763], Section 6.
Jakab & Brockners Expires 1 May 2027 [Page 16]
Internet-Draft Zero-Configuration Agent Discovery October 2026
+=================+=============+============================+
| Key | Presence | Value |
+=================+=============+============================+
| txtvers | always, | Version of the TXT record |
| | first | format, 1 |
+-----------------+-------------+----------------------------+
| proto | always | Agent protocol identifier; |
| | | fixed value a2a |
+-----------------+-------------+----------------------------+
| desc-ref | recommended | Reference to an external |
| | | descriptor of the agent, |
| | | shared by its instances |
+-----------------+-------------+----------------------------+
| desc-ref-digest | recommended | Algorithm-prefixed digest |
| | | of the AgentCard, such as |
| | | sha-256:... |
+-----------------+-------------+----------------------------+
| proto-ver | optional | An A2A protocol version |
| | | the agent supports |
+-----------------+-------------+----------------------------+
| well-known | when needed | AgentCard suffix under |
| | | /.well-known/; omitted for |
| | | agent-card.json |
+-----------------+-------------+----------------------------+
| title | when needed | AgentCard name, when the |
| | | instance name differs from |
| | | it |
+-----------------+-------------+----------------------------+
| alpn | optional | HTTP versions the endpoint |
| | | supports, such as h2,h3 |
+-----------------+-------------+----------------------------+
Table 2: TXT record keys for an A2A agent under _agents._tcp
The keys, their values, and their requirement levels are defined in
[I-D.blwm-dawn-common-discovery-parameters]. This document always
includes "proto", because a discovering party that browses
"_agents._tcp" learns the protocol of each instance only from this
key. When "alpn" is absent, a discovering party uses HTTPS over TCP.
Jakab & Brockners Expires 1 May 2027 [Page 17]
Internet-Draft Zero-Configuration Agent Discovery October 2026
The descriptor is intentionally small. Fields such as the agent
description and version, skills and their tags, authentication
schemes, protocol bindings, and input and output modes are not placed
in DNS-SD records; they are retrieved from the AgentCard.
Capabilities in particular are not advertised: there is no shared
vocabulary for them, and they change more often than DNS records
should. This keeps the advertisement within the size that DNS-SD
carries efficiently (see [RFC6763], Section 6.2, on TXT record
sizing) and avoids exposing more agent metadata than enumeration
requires.
A discovering party can encounter advertisements from independent
implementations. It should ignore TXT keys it does not recognize and
should tolerate the absence of optional keys. Additional keys may be
present; this document does not describe any beyond those above.
As in [I-D.blwm-dawn-common-discovery-parameters], the key that names
the agent protocol ("proto") is distinct from "alpn", which carries
only the HTTP versions an endpoint supports and never names the agent
protocol. No key names a protocol binding or transport; for A2A,
these are learned from the AgentCard.
6.1.5. Example
The following shows the DNS-SD records for an A2A room scheduling
agent named "sched", running on host "room1.local" port 8002. Its
AgentCard is published at the A2A well-known path
"https://room1.local:8002/.well-known/agent-card.json", so the "well-
known" key is omitted. It carries the descriptor reference
"urn:air:example.com:rooms:sched", supports A2A protocol version
"1.0", and advertises the digest of its current AgentCard.
_agents._tcp.local. PTR sched._agents._tcp.local.
_a2a._sub._agents._tcp.local. PTR sched._agents._tcp.local.
sched._agents._tcp.local. SRV 0 0 8002 room1.local.
sched._agents._tcp.local. TXT (
"txtvers=1"
"proto=a2a"
"proto-ver=1.0"
"desc-ref=urn:air:example.com:rooms:sched"
"desc-ref-digest=sha-256:<base64url-digest>"
)
A discovering party that browses "_agents._tcp.local." or
"_a2a._sub._agents._tcp.local." learns of the "sched" instance and
sees from "proto=a2a" that it speaks A2A. If the agent is of
interest, the party reconstructs the AgentCard URL from the SRV host
and port and the default well-known path (no "well-known" key is
Jakab & Brockners Expires 1 May 2027 [Page 18]
Internet-Draft Zero-Configuration Agent Discovery October 2026
present). If it already holds an AgentCard whose digest matches
"desc-ref-digest", it can reuse that AgentCard; otherwise it
retrieves the full AgentCard. It then applies the A2A interface-
selection rules to the AgentCard's supportedInterfaces list before
beginning an A2A session.
For illustration, the same instance can be advertised with SVCB
(Section 5.1.1). An SVCB record at the service instance name takes
the place of both the SRV record, through its TargetName and "port"
parameter, and the TXT record, through its SvcParams. The "txtvers"
key has no SVCB counterpart, and "alpn" is required in SVCB
([I-D.blwm-dawn-common-discovery-parameters]). A deployment that
also serves discovering parties without SVCB support keeps the SRV
and TXT records alongside the SVCB record.
_agents._tcp.local. PTR sched._agents._tcp.local.
_a2a._sub._agents._tcp.local. PTR sched._agents._tcp.local.
sched._agents._tcp.local. SVCB 1 room1.local. (
port=8002
alpn=h2
proto=a2a
proto-ver=1.0
desc-ref=urn:air:example.com:rooms:sched
desc-ref-digest=sha-256:<base64url-digest>
)
6.2. MCP
This section instantiates the model of Section 5.1 for the Model
Context Protocol (MCP) [MCP]. It is descriptive and does not define
a required MCP encoding.
MCP differs from A2A in a way that shapes the instantiation. MCP
defines two standard transports: a "stdio" transport, in which a
client launches the server as a local subprocess, and a "Streamable
HTTP" transport, in which the server exposes a single HTTP endpoint
(the "MCP endpoint") that accepts client connections. A client must
know which transport a server uses before it can connect; the
transport is a property of the deployed server, not something
negotiated once connected. The "stdio" transport has no network
rendezvous and is out of scope for network discovery; only an MCP
server reachable over the Streamable HTTP transport is meaningfully
advertised on a local link. Furthermore, core MCP has no static,
fetchable description document analogous to the A2A AgentCard. A
client learns the server's identity, protocol version, and
capabilities by connecting to the MCP endpoint and performing the MCP
"initialize" handshake, in which client and server negotiate the
protocol version, followed as needed by the "tools/list", "resources/
Jakab & Brockners Expires 1 May 2027 [Page 19]
Internet-Draft Zero-Configuration Agent Discovery October 2026
list", and "prompts/list" operations. The second stage (Section 5.2)
for MCP is therefore a live handshake against the endpoint, not the
retrieval of a document.
6.2.1. Service Type and Protocol Identifier
An MCP server that participates in local discovery is advertised as a
DNS-SD service instance of the protocol-independent service type
"_agents._tcp" (Section 5.1), in the "local." domain when using mDNS
[RFC6762], and of its "_mcp" subtype. It is identified as an MCP
server by the descriptor key "proto=mcp" in its TXT record (see
Section 6.2.4). No MCP-specific service type is defined; a
discovering party distinguishes MCP servers from agents of other
protocols by the "proto" key, or enumerates only MCP servers by
browsing "_mcp._sub._agents._tcp". Subtype names are scoped to their
parent service type ([RFC6763], Section 16), so this subtype does not
conflict with the unrelated "mcp" entry in the Service Name registry
[RFC6335].
6.2.2. MCP-to-DNS-SD Mapping
The descriptor projects a minimal subset of the MCP server's
publication information onto a DNS-SD service instance. It carries
only what is useful for enumeration and lightweight selection; the
authoritative identity, protocol version, and capability set are
obtained by connecting to the MCP endpoint and performing the
"initialize" handshake (see Section 5.2). The mapping described in
this document is:
Jakab & Brockners Expires 1 May 2027 [Page 20]
Internet-Draft Zero-Configuration Agent Discovery October 2026
+==============+================+===================================+
| Source field | DNS-SD record | Transformation |
| or value | | |
+==============+================+===================================+
| (protocol | TXT key proto, | Fixed value mcp, marking the |
| identifier) | subtype _mcp | instance as an MCP server |
+--------------+----------------+-----------------------------------+
| server title | Service | Encoded as the instance |
| or name | instance name | label (see Section 6.1.3) |
+--------------+----------------+-----------------------------------+
| server title | TXT key title | Copied when the instance |
| or name | | name differs from it |
+--------------+----------------+-----------------------------------+
| MCP endpoint | SRV target and | The MCP endpoint URL |
| authority | port | authority is parsed into |
| | | host name and port |
+--------------+----------------+-----------------------------------+
| MCP endpoint | TXT key path | Path of the MCP endpoint |
| URL path | | URL; always present, because |
| | | MCP defines no default |
+--------------+----------------+-----------------------------------+
| (external | TXT key desc- | Reference to an external |
| descriptor) | ref | descriptor of the agent, as |
| | | for A2A |
+--------------+----------------+-----------------------------------+
| MCP protocol | TXT key proto- | An MCP protocol version the |
| version | ver | server supports, as an |
| | | optional selection hint |
+--------------+----------------+-----------------------------------+
| all other | (not | Obtained from the |
| information | advertised) | "initialize" handshake at |
| | | the MCP endpoint |
+--------------+----------------+-----------------------------------+
Table 3
Jakab & Brockners Expires 1 May 2027 [Page 21]
Internet-Draft Zero-Configuration Agent Discovery October 2026
For the SRV record, the advertising implementation uses the host name
and port of the MCP endpoint. Following [RFC6763], Section 6.3, the
host name and port are not duplicated in the TXT record. The MCP
Streamable HTTP transport requires a server to provide a single HTTP
endpoint path, but does not define a default one, so the "path" key,
which carries the path component of the MCP endpoint URL, is always
included. A discovering party reconstructs the MCP endpoint URL by
combining the SRV target host name and port with the "path" value,
using the "https" scheme. For example, an SRV record giving host
"room1.local" and port 8003 together with "path=/api/mcp" yields the
MCP endpoint URL "https://room1.local:8003/api/mcp". If the port is
the default for the scheme, an implementation may omit it from the
reconstructed URL.
No transport key is advertised. The transport is a property of the
deployed MCP server that a client needs to know before connecting,
because MCP does not negotiate it once connected. The Streamable
HTTP transport is, however, the only standard MCP transport that is
reachable over the network (see above), so a discovering party uses
Streamable HTTP for every MCP server it discovers through this
mechanism.
The optional "proto-ver" key is a lightweight selection hint only.
It carries an MCP protocol version the server supports, in the MCP
"YYYY-MM-DD" form. A discovering party still connects to the MCP
endpoint and negotiates the protocol version through the MCP
"initialize" handshake, which remains authoritative, so a discovering
party does not rely on "proto-ver" in place of negotiation.
6.2.3. Service Instance Name Encoding and Conflicts
The MCP server's display name is used as the initial candidate for
the "Instance" portion of the DNS-SD service instance name: the title
in the Implementation information the server reports in the
"initialize" handshake, or its name when it has no title. The MCP
specification intends name for programmatic or logical use and title
for user interfaces, and an instance name is meant to be user-visible
([RFC6763], Section 4.1.1), so title is the better source. The
result is encoded and conflict-resolved exactly as described for A2A
in Section 6.1.3. When the instance name differs from the server's
display name, the advertising implementation can include the display
name in the "title" key, as for A2A. As with A2A, a shared "desc-
ref" value, where present, rather than the instance name or the
"title" key, lets a discovering party recognize the agent if the DNS-
SD instance name changes after conflict resolution, and the name and
title reported in the "initialize" handshake are authoritative.
Jakab & Brockners Expires 1 May 2027 [Page 22]
Internet-Draft Zero-Configuration Agent Discovery October 2026
6.2.4. TXT Record Keys
The TXT record carries the key/value pairs in Table 4, encoded per
[RFC6763], Section 6.
+===========+===============+=======================================+
| Key | Presence | Value |
+===========+===============+=======================================+
| txtvers | always, | Version of the TXT record format, 1 |
| | first | |
+-----------+---------------+---------------------------------------+
| proto | always | Agent protocol identifier; fixed |
| | | value mcp |
+-----------+---------------+---------------------------------------+
| path | always | Path of the MCP endpoint URL; host |
| | | and port from SRV |
+-----------+---------------+---------------------------------------+
| title | when needed | The server's display name, when the |
| | | instance name differs from it |
+-----------+---------------+---------------------------------------+
| desc-ref | recommended | Reference to an external descriptor |
| | | of the agent, shared by its instances |
+-----------+---------------+---------------------------------------+
| proto-ver | optional | An MCP protocol version the server |
| | | supports, in YYYY-MM-DD form |
+-----------+---------------+---------------------------------------+
| alpn | optional | HTTP versions the endpoint supports, |
| | | such as h2,h3 |
+-----------+---------------+---------------------------------------+
Table 4: TXT record keys for an MCP server under _agents._tcp
The "proto" and "desc-ref" keys carry the same protocol-independent
meaning as in the A2A mapping (Section 6.1.4); "path", combined with
the SRV host and port, gives the MCP endpoint, the MCP second-stage
rendezvous point. It is analogous in role to the A2A AgentCard
location, but the endpoint is connected to and handshaked with rather
than retrieved as a document. Because core MCP has no description
document, the MCP descriptor carries neither "well-known" nor "desc-
ref-digest".
This mapping advertises no capability hint, for the same reasons as
for A2A (Section 6.1.4). An MCP server's capabilities, and its tool,
resource, and prompt lists, are dynamic and are obtained
authoritatively from the "initialize" handshake and the "tools/list",
"resources/list", and "prompts/list" operations in the second stage;
a discovering party therefore enumerates MCP capabilities after
connecting rather than from the descriptor.
Jakab & Brockners Expires 1 May 2027 [Page 23]
Internet-Draft Zero-Configuration Agent Discovery October 2026
A discovering party ignores TXT keys it does not recognize and
tolerates the absence of optional keys, as described in
Section 6.1.4.
6.2.5. Example
The following shows the DNS-SD records for an MCP server named
"notes", reachable at the Streamable HTTP endpoint
"https://room1.local:8003/mcp" on host "room1.local" port 8003. It
supports MCP protocol version "2025-11-25".
_agents._tcp.local. PTR notes._agents._tcp.local.
_mcp._sub._agents._tcp.local. PTR notes._agents._tcp.local.
notes._agents._tcp.local. SRV 0 0 8003 room1.local.
notes._agents._tcp.local. TXT (
"txtvers=1"
"proto=mcp"
"path=/mcp"
"proto-ver=2025-11-25"
"desc-ref=urn:air:example.com:rooms:notes"
)
A discovering party that browses "_agents._tcp.local." or
"_mcp._sub._agents._tcp.local." learns of the "notes" instance, sees
from "proto=mcp" that it speaks MCP, reconstructs the MCP endpoint
URL from the SRV host and port and the "path" value, connects using
the Streamable HTTP transport, and performs the MCP "initialize"
handshake to learn the server's authoritative identity, protocol
version, and capabilities before using it.
6.3. Other Instantiations
Instantiations of the model (Section 5.1) for agent protocols other
than A2A and MCP are left for future work. Under the architecture
described in this document, such an instantiation would use the
shared "_agents._tcp" service type, define a value for the "proto"
TXT key and the corresponding subtype, use the common parameters of
[I-D.blwm-dawn-common-discovery-parameters] as lightweight selection
hints, and identify how the protocol's full agent description is
obtained in the second stage (Section 5.2). Per-protocol browsing
uses a subtype of the shared service type; a per-protocol DNS-SD
service type would be a different discovery architecture, not an
instantiation of the shared-service-type model described here.
Jakab & Brockners Expires 1 May 2027 [Page 24]
Internet-Draft Zero-Configuration Agent Discovery October 2026
7. Deployment Considerations
mDNS is, by design, limited to the local link [RFC6762]. On its own,
the approach described here therefore enumerates only agents that are
present on the same link as the discovering party. Section 7.1
describes how managed networks extend it beyond the link.
Because the second stage uses the agent's own transport and
authentication, the reachability of an agent's endpoint is
independent of whether the agent was discovered through DNS-SD.
Conversely, successful discovery does not imply that the discovered
endpoint is reachable, authorized for use, or trusted.
Managed networks may filter or constrain multicast traffic.
Deployments that rely on DNS-SD over mDNS need to account for local
network policy, interface selection, segmentation, and any service-
routing or filtering functions that are present in the network.
7.1. Discovery Beyond the Local Link
Managed networks, such as enterprise campus and branch networks,
typically span many links, and the agents that a discovering party is
interested in are not necessarily on its own link. Such networks can
extend DNS-SD over mDNS beyond the link with network functions, such
as mDNS gateways and proxies, that learn the advertisements on some
links and make them available on others under central administration.
The Discovery Proxy [RFC8766] is a standardized example: it answers
unicast DNS-SD queries with the results of mDNS queries on the links
it serves. Agents keep advertising over mDNS. These functions relay
the agents' DNS-SD records, with the TXT record unchanged, so the
descriptor and the two-stage flow are the same as on a single link.
They also give the network operator a central point for visibility
into the agents present in the network, and for applying local policy
(REQ-10). Agent discovery in unicast DNS uses the same parameters
[I-D.blwm-dawn-common-discovery-parameters], so a discovering party
can interpret local and global advertisements in the same way.
In networks where multicast is constrained, an agent can also
register its DNS-SD records by unicast, with the Service Registration
Protocol (SRP) [RFC9665]. SRP complements mDNS rather than replacing
it: an agent that registers with SRP can keep advertising over mDNS,
so discovery on its own link does not depend on the SRP registrar.
The agent sends the same PTR, SRV, and TXT records as over mDNS,
including the per-protocol subtype, in a single update to the
registrar. The registrar publishes them in a unicast DNS domain for
the lifetime of a lease, which the agent renews. Following
[RFC9665], Section 6.3, that domain is preferably a dedicated
subdomain, such as "dnssd.example.com", rather than the
Jakab & Brockners Expires 1 May 2027 [Page 25]
Internet-Draft Zero-Configuration Agent Discovery October 2026
organization's main zone. A discovering party that learns the
domain, for example through DNS-SD domain enumeration ([RFC6763],
Section 11), browses it with ordinary unicast DNS queries, and the
descriptor and the two-stage flow are again unchanged. Like a
gateway, the registrar gives the network operator a central point for
visibility and local policy.
8. Related work
8.1. LAD-A2A project
The LAD-A2A project [LAD-A2A] also advertises A2A agents on the local
link over mDNS, and likewise positions local discovery as a bootstrap
that precedes agent interaction. The two efforts are similar in
objective and approach, but differ in emphasis:
* LAD-A2A is broader in scope and describes the use of mDNS, well-
known endpoints, and DHCP options for discovery. It also
concentrates on application-layer trust, including TLS, signed
AgentCards, decentralized identifiers, and explicit user consent.
It uses an A2A-specific service type ("_a2a._tcp"), whereas this
document uses the protocol-independent "_agents._tcp" service type
with an "_a2a" subtype.
* This document focuses on local agent discovery as a protocol-
independent model (Section 5.1), using a single service name,
"agents", across agent protocols with A2A as one instantiation,
and describes a concrete AgentCard-to-DNS-SD field mapping and
descriptor (Section 6.1.2, Section 6.1.4).
The two approaches can coexist on the same link because DNS-SD
service types are independent. An implementation that wants to be
discoverable by clients following both approaches could advertise
both "_agents._tcp" with proto=a2a and the A2A-specific service type
used by LAD-A2A, but such dual advertisement is outside the scope of
this document.
8.2. MCP DNS discovery
[I-D.morrison-mcp-dns-discovery] describes a DNS-based first-stage
rendezvous for Model Context Protocol servers, carried in unicast DNS
TXT records at an underscore-prefixed label in the domain's own zone.
It performs the same two-stage function as the mechanism described
here, with domain rather than link scope. DNS-SD over mDNS
enumerates the agents present on a link, whereas unicast DNS
enumerates the agents published under a domain, so the two are
complementary. The two-stage rendezvous abstraction (a lightweight
step that locates a candidate, identifies its protocol family, and
Jakab & Brockners Expires 1 May 2027 [Page 26]
Internet-Draft Zero-Configuration Agent Discovery October 2026
points to an authoritative description) is not inherently link-
scoped; this document instantiates it for the local link, while
unicast DNS provides the analogous function at domain scope.
8.3. OpenClaw project
OpenClaw [OpenClaw] is an existing deployment of the same pattern.
Its gateway advertises a per-application DNS-SD service type
(_openclaw-gw._tcp) over mDNS on the local link, carrying a small TXT
descriptor of non-secret hints (a role marker, display name, endpoint
and port, and a TLS fingerprint); clients, including the OpenClaw iOS
and Android companion applications, browse for it to find and connect
to a nearby gateway. OpenClaw also offers a wide-area mode using
unicast DNS-SD for cross-network discovery, consistent with
Section 7. Notably, its own guidance is that the TXT records are
unauthenticated and that clients should rely on the resolved endpoint
rather than trusting TXT hints, the same posture taken in Section 9
of this document. OpenClaw is thus an independent instance of the
general model of Section 5.1, arrived at for a different agent
system, which illustrates that agent vendors are adopting mDNS/DNS-SD
for the local-discovery step.
9. Security Considerations
mDNS and DNS-SD do not authenticate advertisements. On a shared
local link, any host can publish an agent service instance, including
one that impersonates another agent, and can populate the TXT keys
with arbitrary values. The records described in this document,
including the "desc-ref" and "title" keys and the reconstructed
rendezvous location (the SRV host and port together with a "well-
known" or "path" key), are therefore unauthenticated hints and cannot
be treated as proof of an agent's identity. In particular, "desc-
ref" refers to a descriptor of the agent without proving who operates
it, and on the local link "desc-ref-digest" is only a freshness
signal: because the advertisement is not authenticated, it provides
no integrity protection for the description it refers to. Likewise,
"title" is a display name chosen by the advertiser, which can imitate
another agent, so a discovering party presents it as an unverified
name. A per-protocol subtype narrows browse results but provides no
confidentiality; the advertisement remains visible to every host on
the link.
Jakab & Brockners Expires 1 May 2027 [Page 27]
Internet-Draft Zero-Configuration Agent Discovery October 2026
The same holds beyond the link (Section 7.1). A gateway or proxy
relays advertisements without making them more trustworthy. SRP's
first-come, first-served naming ([RFC9665], Section 3.2.4.1) only
prevents another agent from taking over a name that is already
registered; it does not establish who operates the agent that
registered it. A registrar that accepts registrations only from
within its administrative domain ([RFC9665], Section 6.1) limits who
can register, but does not vouch for what they register.
Trust is established in the second stage, not the first. A
discovering party should treat the DNS-SD records only as a means of
locating candidate agents, and should authenticate the agent when it
connects, using the transport security and authentication mechanisms
of the agent protocol. For A2A, this could include TLS server
authentication and the authentication schemes declared in the
retrieved AgentCard. For MCP, this could include TLS server
authentication of the MCP endpoint and any authorization the server
requires during or after the "initialize" handshake. In particular,
optional selection hints, such as "proto-ver", are advisory and can
be forged. They are not a basis for any security decision.
Because the SRV host and port and a "well-known" or "path" key cause
a discovering party to reconstruct and contact a URL learned from an
unauthenticated multicast advertisement, implementations are advised
to apply the same caution they would apply to any externally supplied
URL. This includes applying local policy to accepted schemes,
address ranges, ports, redirects, credentials, and cross-domain
requests. In particular, the practice in this document of defaulting
the scheme to "https" and, when the "well-known" key is absent,
defaulting to the A2A well-known path, does not by itself provide any
security guarantee; the reconstructed endpoint is still attacker-
influenced and is authenticated only in the second stage.
For MCP specifically, connecting to a discovered endpoint means
initiating an MCP session with a server chosen from an
unauthenticated advertisement. An MCP client should apply its usual
protections for connecting to an untrusted MCP server, including
validating the endpoint's TLS identity before trusting it, and should
not treat tools, resources, or prompts subsequently enumerated from
that server as trustworthy solely because the server was discovered
on the local link. The MCP transport-security guidance, such as
validating the "Origin" header and binding local servers
appropriately, continues to apply and is independent of how the
server was discovered.
Advertising agents expose their existence, and any metadata included
in their advertisements, to other hosts on the link and, when their
advertisements are made visible beyond the link (Section 7.1), to the
Jakab & Brockners Expires 1 May 2027 [Page 28]
Internet-Draft Zero-Configuration Agent Discovery October 2026
hosts that can discover them there. Advertising implementations
should therefore advertise by default only the keys that a
discovering party needs to select and reach the agent, and provide a
setting for which optional keys, such as "title", are included, as
[OpenClaw] does with its discovery modes. Network operators who
consider the set of agents present in their network, their names, or
the protocols they speak to be sensitive should account for this
exposure and apply local policy (REQ-10).
General mDNS and DNS-SD security considerations apply and are
discussed in [RFC6762] and [RFC6763].
10. IANA Considerations
IANA is requested to register the following service name in the
"Service Name and Transport Protocol Port Number Registry" [RFC6335]:
Service Name: agents
Transport Protocol(s): TCP, UDP
Assignee: IESG (iesg@ietf.org)
Contact: IETF Chair (chair@ietf.org)
Description: Discovery of AI agents
Reference: This document. Uses IP multicast when advertised with
mDNS [RFC6762].
Port Number: None
Assignment Notes: Defined subtypes: "_a2a", "_mcp". Defined TXT
keys: see [I-D.blwm-dawn-common-discovery-parameters].
No port number is needed, because the port is conveyed per instance
in the SRV record (see Section 6.1.2). This document uses
"_agents._tcp" (Section 5.1). The "udp" registration accommodates
agents that are reachable only over a transport other than TCP, such
as QUIC (Section 5.1). The subtypes are recorded in the notes field,
following [RFC6763], Section 16. Because "_tcp" and "_udp" are
already registered in the "Underscored and Globally Scoped DNS Node
Names" registry [RFC8552], no registration is needed there. A
registry for "proto" key values (the per-protocol agent identifiers,
such as "a2a" and "mcp") is a matter for
[I-D.blwm-dawn-common-discovery-parameters]; this document does not
define one.
Jakab & Brockners Expires 1 May 2027 [Page 29]
Internet-Draft Zero-Configuration Agent Discovery October 2026
11. Acknowledgements
The authors would like to thank Blake Morrison, Chris Hood, Jim
Mozley, Michael Richardson, and Nic Williams for their feedback.
12. References
12.1. Normative References
[I-D.blwm-dawn-common-discovery-parameters]
Brockners, F., Jakab, L., Williams, N., and J. Mozley,
"Common Agent DNS Discovery Parameters", Work in Progress,
Internet-Draft, draft-blwm-dawn-common-discovery-
parameters-00, 27 October 2026,
<https://datatracker.ietf.org/doc/html/draft-blwm-dawn-
common-discovery-parameters-00>.
[RFC6335] Cotton, M., Eggert, L., Touch, J., Westerlund, M., and S.
Cheshire, "Internet Assigned Numbers Authority (IANA)
Procedures for the Management of the Service Name and
Transport Protocol Port Number Registry", BCP 165,
RFC 6335, DOI 10.17487/RFC6335, August 2011,
<https://www.rfc-editor.org/rfc/rfc6335>.
[RFC6762] Cheshire, S. and M. Krochmal, "Multicast DNS", RFC 6762,
DOI 10.17487/RFC6762, February 2013,
<https://www.rfc-editor.org/rfc/rfc6762>.
[RFC6763] Cheshire, S. and M. Krochmal, "DNS-Based Service
Discovery", RFC 6763, DOI 10.17487/RFC6763, February 2013,
<https://www.rfc-editor.org/rfc/rfc6763>.
12.2. Informative References
[A2A] Linux Foundation, "Agent2Agent (A2A) Protocol
Specification, Version 1.0.1", 26 May 2026,
<https://github.com/a2aproject/A2A/blob/v1.0.1/docs/
specification.md>.
[I-D.farrel-dawn-terminology]
Farrel, A., Yao, K., Schott, R., and N. Williams,
"Terminology for the Discovery of Agents With Names
(DAWN)", Work in Progress, Internet-Draft, draft-farrel-
dawn-terminology-05, 19 October 2026,
<https://datatracker.ietf.org/doc/html/draft-farrel-dawn-
terminology-05>.
Jakab & Brockners Expires 1 May 2027 [Page 30]
Internet-Draft Zero-Configuration Agent Discovery October 2026
[I-D.gakiwate-dnssd-use-svcb]
Akiwate, G. and T. Pauly, "Use SVCB with DNS Service
Discovery", Work in Progress, Internet-Draft, draft-
gakiwate-dnssd-use-svcb-00, 28 February 2025,
<https://datatracker.ietf.org/doc/html/draft-gakiwate-
dnssd-use-svcb-00>.
[I-D.king-dawn-requirements]
King, D. and A. Farrel, "Requirements for the Discovery of
Agents With Names (DAWN)", Work in Progress, Internet-
Draft, draft-king-dawn-requirements-02, 19 October 2026,
<https://datatracker.ietf.org/doc/html/draft-king-dawn-
requirements-02>.
[I-D.morrison-mcp-dns-discovery]
Morrison, B., "Discovery of Model Context Protocol Servers
via DNS TXT Records", Work in Progress, Internet-Draft,
draft-morrison-mcp-dns-discovery-09, 21 October 2026,
<https://datatracker.ietf.org/doc/html/draft-morrison-mcp-
dns-discovery-09>.
[I-D.mozleywilliams-dnsop-dnsaid]
Mozley, J., Williams, N., Sarikaya, B., Schott, R., and J.
Damick, "DNS for AI Discovery", Work in Progress,
Internet-Draft, draft-mozleywilliams-dnsop-dnsaid-02, 27
May 2026, <https://datatracker.ietf.org/doc/html/draft-
mozleywilliams-dnsop-dnsaid-02>.
[LAD-A2A] LAD-A2A Contributors, "LAD-A2A: Local Agent Discovery for
A2A", 2025, <https://lad-a2a.org/>.
[MCP] MCP Community, "Model Context Protocol Specification,
Revision 2025-11-25", 25 November 2025,
<https://modelcontextprotocol.io/
specification/2025-11-25/>.
[OpenClaw] OpenClaw, "OpenClaw Gateway: Bonjour discovery", 2026,
<https://docs.openclaw.ai/gateway/bonjour>.
[RFC8552] Crocker, D., "Scoped Interpretation of DNS Resource
Records through "Underscored" Naming of Attribute Leaves",
BCP 222, RFC 8552, DOI 10.17487/RFC8552, March 2019,
<https://www.rfc-editor.org/rfc/rfc8552>.
[RFC8766] Cheshire, S., "Discovery Proxy for Multicast DNS-Based
Service Discovery", RFC 8766, DOI 10.17487/RFC8766, June
2020, <https://www.rfc-editor.org/rfc/rfc8766>.
Jakab & Brockners Expires 1 May 2027 [Page 31]
Internet-Draft Zero-Configuration Agent Discovery October 2026
[RFC9460] Schwartz, B., Bishop, M., and E. Nygren, "Service Binding
and Parameter Specification via the DNS (SVCB and HTTPS
Resource Records)", RFC 9460, DOI 10.17487/RFC9460,
November 2023, <https://www.rfc-editor.org/rfc/rfc9460>.
[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/rfc/rfc9665>.
Appendix A. Changes Since Previous Revisions
This section is to be removed before publishing as an RFC.
A.1. Since -00
* Uses the common discovery parameters of
[I-D.blwm-dawn-common-discovery-parameters], which are shared with
agent discovery in unicast DNS, instead of defining its own keys.
- Reworks the A2A keys: a2aProtoVer becomes proto-ver; the
AgentCard location moves from the full URL in cardUrl to the
suffix in well-known; and the AgentCard name moves from agentId
to title, which is for display only and is carried only when
the instance name differs from the name.
- Drops the A2A keys caps, ver (the agent's own version), and
a2aBinding, and adds desc-ref and desc-ref-digest.
* Renames the service type from _agent._tcp to _agents._tcp, and
adds the _a2a and _mcp subtypes for browsing one protocol.
* Advertises an agent that offers several protocols as one instance
per protocol, linked by desc-ref, with exactly one TXT record per
instance, beginning with txtvers.
* Points the A2A SRV record at the host that serves the AgentCard;
the endpoint URLs now come from the AgentCard, and host and port
are no longer repeated in TXT.
* Adds an MCP instantiation, and generalizes the second stage to
cover both retrieving a description document and performing a
native handshake.
* Specifies how a name that cannot be a DNS-SD instance name is
handled, and takes MCP instance names from the server's title.
Jakab & Brockners Expires 1 May 2027 [Page 32]
Internet-Draft Zero-Configuration Agent Discovery October 2026
* Adds a use case for visitors in shops, venues, and offices,
describes the target networks as ranging from a single link to a
site, and points to the stadium use case of the common parameters
draft.
* Describes discovery beyond the local link, through mDNS gateways
and proxies, such as the Discovery Proxy [RFC8766], and optionally
through registration with SRP [RFC9665], including how instance
names behave there; -00 left discovery across links out of scope.
* States that selection by capability is left to the second stage,
and why.
* Relates the descriptor to the Minimum Discoverable Information of
[I-D.farrel-dawn-terminology], adds MCP DNS discovery
[I-D.morrison-mcp-dns-discovery] to the related work, and extends
the Security Considerations to MCP, desc-ref, desc-ref-digest,
title, subtypes, and discovery beyond the link, with advertising
implementations, rather than operators, keeping advertisements
minimal by default.
* Requests registration of the agents service name for TCP and UDP,
instead of making no request of IANA.
* Adds Acknowledgements, and Toerless Eckert as a contributor.
Contributors
Alberto Rodriguez-Natal
Cisco Systems
Email: natal@cisco.com
Eric Voit
Cisco Systems
Email: evoit@cisco.com
Pradeep Kathail
Cisco Systems
Email: pkathail@cisco.com
Toerless Eckert
Futurewei USA
Email: tte@cs.fau.de
Jakab & Brockners Expires 1 May 2027 [Page 33]
Internet-Draft Zero-Configuration Agent Discovery October 2026
Authors' Addresses
Lorand Jakab
Cisco Systems
Email: lojakab@cisco.com
Frank Brockners
Cisco Systems
Email: fbrockne@cisco.com
Jakab & Brockners Expires 1 May 2027 [Page 34]