Agent Registration and Discovery Protocol ARDP · /.well-known/ardp/meta

Overview

It solves agent liveness and endpoint churn using a TTL-based model: an agent registers with a server, the server assigns and enforces TTL bounds (advertised at /.well-known/ardp/meta, e.g. ttl_min/ttl_max/auth_profiles/supported_protocols), and clients are expected to refresh at roughly half the TTL with jitter ^[1]. Discovery is split into a RESOLVE operation (map a registered agent identifier to active endpoints and capabilities, access-controlled) and a QUERY operation (discover by capability or namespace, also authorized) ^[1]. Privacy is a first-class concern: QUERY responses are minimized by default to just an identifier and status, with full detail gated behind an explicit detail=full request parameter, and the server must mark a response "redacted": true when fields are withheld ^[1].

ARDP does not define its own transport-security handshake beyond referencing a jws-proof-of-control auth profile in its example metadata, and does not attempt to be a full registry federation protocol -- there is no evidence in the text read of a cross-registry query mechanism comparable to ARD's federation parameter ^[1].

Current state: the IETF Datatracker's own status page (a separate fetch from the draft text itself) shows this document's IESG state as "Expired -- Expired & archived", the sole author's listed contact "may no longer be valid", and no revision beyond -01 ^[1][2]. Phase is recorded as withdrawn on the strength of that authoritative status confirmation, not "unknown": the document is formally lapsed and archived, not merely undocumented. No implementation, deployment, or adopter evidence was located; implementation is recorded as "none found" rather than unknown, since a specific negative search (checking for a follow-up draft and any linked reference implementation) turned up nothing, as distinct from simply not having looked.

What is disputed or unknown: whether this specification is abandoned (an expired, unrenewed single-author draft is a weak signal either way, since many individual IETF drafts lapse and are later resubmitted) or quietly superseded by the author moving to a different venue; and, separately, the confusable-name risk with the unrelated, much more actively developed ARD (Agentic Resource Discovery) specification, which shares three of four words in its expansion and a similar acronym but has no connection in governance, mechanism, or authorship.

Discovery and probe

RESOLVE/QUERY operations over an HTTPS binding, with a minimal metadata resource advertising TTL bounds and supported auth profiles · /.well-known/ardp/meta

GET https://{origin}/.well-known/ardp/meta
parse:     200 response is JSON with ttl_min, ttl_max, auth_profiles and supported_protocols fields
on absent: Domain does not run an ARDP-compliant registration/discovery endpoint

Live check run by the MCP probe: GET /.well-known/ardp/meta. ARDP meta declares TTL bounds and auth profiles.

{
 "path": "/.well-known/ardp/meta",
 "expect": {
  "json_keys": [
   "ttl_min",
   "auth_profiles"
  ]
 },
 "note": "ARDP meta declares TTL bounds and auth profiles."
}

Instances

No instances recorded yet.

See also

• Pairs with: Agent2Agent Protocol (A2A), Model Context Protocol, Agent Name Service v2 (ANS), DNS-AID (DNS for AI Discovery), Agentic Resource Discovery (ARD), A Framework for Agent Discovery in DAWN
• Not to be confused with: ARD (Agentic Resource Discovery, unrelated Google/Microsoft/HF spec with a similar acronym and near-identical name)

References

1. draft-pioli-agent-discovery-00 (February 2026) specifies ARDP, a lightweight protocol providing stable agent identities (AIDs), dynamic endpoint resolution via RESOLVE/QUERY operations, capability advertisement across MCP/A2A/HTTP/gRPC, TTL-bounded presence records, and a minimal /.well-known/ardp/meta resource; a successor revision, draft-pioli-agent-discovery-01, was posted on 2026-02-24 and expired on 2026-08-28. — https://datatracker.ietf.org/doc/html/draft-pioli-agent-discovery-00 (2026-02) VERIFIED
2. IETF Datatracker's status page for draft-pioli-agent-discovery (fetched 2026-09-06, distinct page from the htmlized draft text) shows the document's IESG state as "Expired", type "Expired Internet-Draft (individual) -- Expired & archived", with a single author (Roberto Pioli) whose listed e-mail "may no longer be valid", and only one revision beyond -00 (the -01 text, latest revision dated 2026-02-24) -- independently confirming this is a lapsed, unrenewed single-author submission rather than an active track. — https://datatracker.ietf.org/doc/draft-pioli-agent-discovery/ (2026-09-06) VERIFIED
3. Expired Internet-Draft (expired 2026-08-28) — https://datatracker.ietf.org/doc/draft-pioli-agent-discovery/ (2026-02-24) VERIFIED

Disputed: This draft's own name and acronym (ARDP, Agent Registration and Discovery Protocol) is one character and one word away from the entirely unrelated, much more prominent ARD (Agentic Resource Discovery, ard.md) from Google/Microsoft/Hugging Face. They are independent specifications with no shared authorship or mechanism found in the texts read.

Verification log and errata

• 2026-10-16 — Expired Internet-Draft (expired 2026-08-28)
• 2026-09-06 — verified against draft-pioli-agent-discovery-00 spec text via datatracker.ietf.org; noted the draft as expired (date and successor corrected 2026-10-30: draft-pioli-agent-discovery-01 was posted on 2026-02-24 and expired on 2026-08-28)
• 2026-09-06 — fetched IETF Datatracker status page as second, independent evidence entry confirming Expired/archived state; corrected version/spec_url from -00 to the actual latest revision -01; set phase from unknown to withdrawn on the strength of the datatracker's own Expired status