Agent Surface

Discovery & AEO

DNS Discovery for Agent Services

Advertise agent endpoints through DNS when your clients support the emerging DNS-AID proposal

Last verified 2026-09-24

DNS-AID lets an organization advertise agent services through its domain's DNS records. A compatible client can look up connection details from a known domain, then use the advertised protocol to contact the service.

When to use it

Consider DNS discovery when you operate agent services and an intended client supports it. Keep HTTP discovery through well-known endpoints and catalogs available to other clients.

As verified on 10 September 2026, DNS-AID revision 02 is an active individual Internet-Draft. It is not an RFC or an IETF-endorsed standard. Pin the revision your client implements before publishing records. Draft status.

What to publish

The proposal uses service binding records to describe connections. Agree on the record names, protocol identifiers and parameters with your client implementation; do not assume every client supports every discovery route.

Cloudflare's published implementation guide uses names under _agents, ServiceMode SVCB records or HTTPS records, and connection parameters such as protocol and port. Its scanner probes _index._agents, _a2a._agents and _mcp._agents with both record types, plus a TXT index. These are scanner probes, not a requirement to publish services you do not operate.

Use your DNS provider's record editor or infrastructure configuration. Keep records tied to the same service inventory as your HTTP metadata. Choose a TTL that allows planned endpoint changes, and remove obsolete records when a service is retired.

Verify discovery through to use

  1. Query the names the intended client uses. For example, dig SVCB _mcp._agents.example.com +dnssec inspects one possible entry point; substitute your domain and client's record name.
  2. Check that the client understands the returned target and parameters. An empty answer, unsupported record or parse failure needs an explicit fallback or error.
  3. Validate DNSSEC using a trusted validating resolver. Requesting DNSSEC data with +dnssec alone does not establish that validation succeeded. Cloudflare's guide recommends signing public discovery zones.
  4. Connect to the advertised endpoint, verify TLS, and exercise a read-only operation using the advertised protocol. Continue to enforce the service's authentication and permission requirements.
  5. Repeat with the actual supported client after a DNS or endpoint change. Keep the lookup, resolver, validation result, endpoint response and timestamp as evidence.

DNSSEC authenticates DNS data. It does not establish that a service is safe, authorize a tool call, or replace TLS and application authorization. Treat discovered endpoints as untrusted until the relevant checks succeed.

How to interpret a scanner result

Cloudflare checks DNS-AID independently. Its catalog check also probes _catalog._agents TXT and _search._agents SRV records; those are separate routes within catalog discovery. A scan can show what was found, but cannot establish compatibility with every client.

Record DNS-AID as optional unless your product promises it to a supported consumer. A website that has no such integration does not need DNS records merely to improve a score. See the public scanner comparison for the distinction between coverage and implementation requirements.

On this page