ANP vs ARD/DNS

Two approaches to agent discovery: decentralized identity vs. centralized directory. ANP · AD (ARD) · IETF Draft

ANP Agent Network Protocol

Decentralized agent communication built on W3C DIDs. Each agent has a cryptographic identity (did:wba) anchored to a domain. Discovery is peer-to-peer: agents resolve each other's DID documents directly from their hosting domains.

No central registry. Authentication and capability negotiation happen inline via the meta-protocol layer. Designed for open, cross-organization agent networks.

ARD/DNS Agent Resource Directory

Centralized soft-state registry adapted from CoRE Resource Directory (RFC 9176). Agents register their base URI, protocols, and capabilities via HTTP POST. Clients query a single lookup endpoint to find agents by capability, domain, or protocol.

DNS-SD (_ad._tcp) enables zero-config discovery of the directory itself on local networks. Registrations expire unless refreshed.

Dimension ANP ARD/DNS
Architecture Decentralized, peer-to-peer Centralized directory
Identity DID:WBA (cryptographic, domain-anchored) Agent name + domain (soft-state)
Discovery DID resolution → agent description doc DNS-SD → well-known URI → lookup API
Auth End-to-end encrypted, DID-based Not specified (transport-level)
Capability query Fetch agent description, parse capabilities Query /ad/l?cap_name=X
Stale entries N/A (live DID resolution) TTL-based expiry (default 24h)
Cross-org Native (any domain) Requires federation or shared directory
Local network Requires DNS/HTTPS infrastructure DNS-SD zero-config (_ad._tcp)
Standards basis W3C DID, custom specs RFC 9176 (CoRE RD), RFC 6763 (DNS-SD)
Deployment Per-agent (host a DID document) One directory service per scope

ANP   How an agent finds another agent

1
Agent A knows Agent B's DID: did:wba:example.com:agents:translator
↓
2
Resolve DID → fetch https://example.com/.well-known/did.json (or path-based)
↓
3
DID document contains service endpoints and public keys
↓
4
Fetch agent description doc from service endpoint → capabilities, protocols
↓
5
Negotiate protocol via meta-protocol layer, establish authenticated session

ARD   How an agent finds another agent

1
Agent A needs a translator. Knows (or discovers via DNS-SD) the directory at ad.example.com
↓
2
Confirm directory: GET /.well-known/ad → registration + lookup endpoints
↓
3
Query: GET /ad/l?cap_name=translate&protocol=a2a
↓
4
Directory returns matching agents with base URIs and capability metadata
↓
5
Agent A contacts Agent B directly at its base URI using the negotiated protocol

ARD   DNS-SD local discovery

1
Browse local network: _ad._tcp.local (mDNS/DNS-SD)
↓
2
Resolve SRV + TXT records → directory host and port
↓
3
Continue with ARD flow above (steps 2–5)

ANP   Strengths

  • No single point of failure or control
  • Cryptographic identity — hard to spoof
  • Works across organizations without coordination
  • End-to-end encryption built in
  • Agent identity persists across directory changes

Weaknesses

  • Requires DNS + HTTPS for every agent
  • No capability-based search without fetching each agent
  • DID resolution adds latency
  • Complex key management
  • Not suited for local/offline networks

ARD/DNS   Strengths

  • Single query to find agents by capability
  • DNS-SD works on local networks, no internet needed
  • Simple HTTP API, easy to implement
  • Grounded in proven standards (RFC 9176, RFC 6763)
  • Soft-state TTL handles stale entries automatically

Weaknesses

  • Central directory is a single point of failure
  • No built-in authentication or identity verification
  • Cross-org discovery requires federation
  • Agents must actively refresh registrations
  • Directory operator can censor or manipulate listings
Enterprise multi-agent system (single org)
ARD wins
A company runs 50 internal agents. Capability-based lookup via a central directory is simpler to operate and query. DNS-SD handles local deployment. No need for cryptographic identity across org boundaries.
Cross-organization agent collaboration
ANP wins
Agent at company A needs to invoke an agent at company B. ANP's DID-based identity works without a shared directory. Each agent is discoverable from its domain, and authentication is cryptographic.
IoT / local network agent mesh
ARD wins
Agents on a factory floor or home network. DNS-SD (_ad._tcp) enables zero-config directory discovery without internet access. ANP requires DNS and HTTPS infrastructure that may not exist locally.
Open internet agent marketplace
ANP wins
Thousands of agents from different vendors. No central authority should control the registry. ANP's decentralized model scales without coordination. ARD would require a trusted global directory operator.
Capability-based agent search
ARD wins
"Find me any agent that can translate text." ARD's lookup API answers this in one query. ANP has no equivalent — you need to know an agent's DID before you can inspect its capabilities.
High-security agent interactions
ANP wins
Financial or medical agents where identity spoofing is a serious risk. ANP's cryptographic DID-based identity and end-to-end encryption provide stronger guarantees than ARD's name-based soft-state registration.
Hybrid: discovery + secure interaction
Complementary
Use ARD for capability-based discovery (find the right agent), then use ANP for authenticated, encrypted communication once the agent is found. The AD draft explicitly notes complementarity with A2A and MCP interaction protocols.