Update review · November 16, 2026
Zero-Configuration Agent Discovery -01, Now Built on the Common Parameters
What changed in -01?
Almost everything on the wire. The document went from 20 pages to 34.
The -00 review covered a draft that defined its own TXT keys under _agent._tcp, with A2A as the only example. In -01 the service type is _agents._tcp, every A2A key is renamed or gone, and MCP gets its own instantiation.
Nothing a -00 implementation advertises would be understood by a -01 client.
Does -01 defer to the common parameters draft?
Yes, fully. Appendix A.1 says it uses the keys of draft-blwm-dawn-common-discovery-parameters "instead of defining its own keys." Section 6.1.4 says the keys, their values, and their requirement levels are defined there.
That draft is now the only non-RFC normative reference (Section 12.1). Both authors of this draft are also authors of it, alongside two Infoblox authors.
What is left here is the mDNS-specific part: the service type, the subtypes, the two-stage flow, and the A2A and MCP mappings.
What was added and removed
- Subtypes (Section 5.1). A client that only wants A2A browses
_a2a._sub._agents._tcp._mcpworks the same way. - An MCP mapping (Section 6.2). MCP has no description document, so stage two is the
initializehandshake. desc-refanddesc-ref-digest. A reference to an external descriptor, and a digest of the AgentCard so a client can skip re-fetching it.- Discovery beyond the link (Section 7.1). mDNS gateways, the Discovery Proxy (RFC 8766), and SRP (RFC 9665).
- A real IANA request (Section 10). The
agentsservice name for TCP and UDP.
The caps, ver, and a2aBinding keys are gone. Capabilities are now left to stage two (Section 5.2.1). That is the right call.
Was the previous feedback addressed?
Partly addressed:
- No normative language. This document is still Informational and still uses no BCP 14 terms (Section 4). The requirement levels moved to the common parameters draft, which is Standards Track.
- The LAD-A2A fork. Subtypes give protocol-specific browsing, which was the practical argument for
_a2a._tcp. Dual advertisement is still out of scope (Section 8.1). - SVCB. Still in Section 5.1.1, and Section 6.1.5 now adds an SVCB example. The details moved to the common parameters draft.
Not addressed:
- No
protoregistry. Section 10 says this "is a matter for" the common parameters draft. That draft's IANA section (Section 6) is a placeholder. Nobody owns the list ofa2a,mcp, and whatever comes next.
New problems
The scheme is always https. Sections 6.1.2 and 6.2.2 tell the client to rebuild the URL with the https scheme. The examples use room1.local. The draft does not say how a host with a .local name gets a certificate a visiting client will accept. Section 9 puts all trust in stage two, so this is the part that has to work.
The charter asks for integrity. This draft offers none at stage one. The charter says WG work items "will provide integrity of all the discovery information exchanged between the DAWN parties." Section 9 says the records are "unauthenticated hints" and that desc-ref-digest "provides no integrity protection" on the local link. The draft needs to argue that stage-two authentication meets the charter's bar, or say what would.
Background of the authors
The author list did not change. Loránd Jakab and Frank Brockners, both Cisco, wrote -00 and -01.
Toerless Eckert (Futurewei USA) was added as a contributor.
Questions worth raising
- Who registers
protovalues? Both drafts point elsewhere. - How does TLS work for
.localnames, and does stage-two authentication satisfy the charter's integrity requirement? - Should this stay Informational? The charter makes the Protocol deliverable Standards Track, "possibly one per use case," and this reads like the local network protocol document.
- Should the common parameters draft be adopted first? This draft and DNS-AID, which share Thursday's 15 minute slot with it, cannot move without it.