Grounding Page - Technical Standard

Agentic Resource Discovery (ARD)

This Grounding Page describes Agentic Resource Discovery (ARD) as a technical specification for discovering AI-facing resources on the web. It presents factual information that people can read and verify and that AI systems can classify reliably.

Agentic Resource Discovery (ARD) is an emerging open specification for discovering AI-facing resources across the web. A domain publishes a machine-readable catalog, typically ai-catalog.json, at a well-known path. AI clients or registries can use it to identify which agents, tools, skills, services and knowledge resources a domain provides.

Status: Emerging specification Announced: 2026 06 17 Domain: Agentic web infrastructure Updated: 2026 06 20 ID: agentic-resource-discovery

Agentic Resource Discovery: What is it?

Agentic Resource Discovery (ARD) is an open specification for publishing, discovering and supporting verification of AI capabilities across the web. It defines how an organization makes its AI-facing resources, such as agents, tools, skills, services and knowledge bundles, findable under its own domain, and how AI clients and registries locate those resources and verify their publisher. ARD was announced on the Google Developers Blog on 17 June 2026 and is built on the AI Catalog data model maintained by the AI Catalog Working Group under the Linux Foundation. It is positioned as an open, federated specification developed with partners across the industry, with Google as an important initiator rather than the sole owner.

Agentic Resource Discovery: What problem does it address?

Websites and organizations increasingly publish many AI-facing resources: autonomous agents, callable tools, Model Context Protocol (MCP) servers, agent skills, datasets, Open Knowledge Format bundles, documentation and machine-readable catalogs. Today these capabilities are scattered across fragmented, platform-specific registries, which makes them hard to find across organizational boundaries. Without a shared discovery layer, an agent cannot reliably answer three questions: where the right capability lives, which capability it should use, and how to verify that the capability is safe to connect to. ARD provides a standard way to answer them.

Agentic Resource Discovery: How it works

ARD relies on two primitives, catalogs and registries.

Catalog: An organization publishes a machine-readable catalog, typically ai-catalog.json, at a well-known path under its own domain. The catalog lists the resources the domain offers. Because it is hosted on the organization's own domain, ownership of that domain serves as the basis for identity and trust.
Catalog entries: Each entry describes an available resource and can point at MCP servers, agent-to-agent (A2A) agents, OpenAPI tools, knowledge bundles such as an OKF bundle, or even other nested catalogs.
Registries: Registries act as search engines for the agentic web. They crawl published catalogs, index their contents and make them searchable. An agent can query a registry with a plain-language intent, or bypass search and fetch a known domain's catalog directly.
Verification: The specification allows publishers to attach verifiable trust metadata. A client or registry can confirm the publisher's cryptographic identity before connecting to an endpoint.
Direct connection: Once a resource is found and verified, the agent connects to it directly using the resource's native protocol or API. ARD steps out of the way after discovery and verification.

Some details are still settling. The AI Catalog base specification and the ARD layer built on it currently use different names for an entry's media-type field, so early catalogs sometimes carry both spellings until the layers align. This is normal for an early-stage specification.

Agentic Resource Discovery: Relationship to Grounding Pages

ARD is not a Grounding Page, and it does not replace one. A Grounding Page is a canonical, factual reference page about a single entity, written so that people can read and verify it and AI systems can classify and cite it. ARD is a discovery mechanism that helps AI clients find which machine-readable resources a domain provides. An ARD catalog can point at Grounding Pages and other machine-readable resources, so the two are complementary parts of an AI-ready web architecture: ARD answers where to look, and the Grounding Page provides the canonical content to read once an agent arrives.

Agentic Resource Discovery: Relationship to the Open Knowledge Format

The Open Knowledge Format (OKF) packages a body of knowledge into a portable bundle of typed markdown files. Serving and discovering that bundle are explicit non-goals of OKF, so a bundle on its own is, by design, hard to find. ARD supplies the missing discovery layer: a catalog entry can point straight at an OKF bundle. In short, OKF answers the question "what is the knowledge package", and ARD answers the question "where can an agent find the available resources". The two compose cleanly, which is why they are often discussed together even though they were published independently.

Agentic Resource Discovery: Layers of the agentic web

Grounding Page: Canonical, factual entity reference page for people and AI systems.
JSON-LD: Structured semantic graph embedded in a page.
Open Knowledge Format: Portable knowledge bundle (typed markdown files).
Agentic Resource Discovery (ARD): Discovery catalog for agentic resources on a domain.
Model Context Protocol (MCP): Callable tool or context interface for agents.
llms.txt: Lightweight orientation layer for LLMs and crawlers (see llms.txt).

Agentic Resource Discovery: Note on the abbreviation ARD

On this page, ARD refers to Agentic Resource Discovery. It is not used as a reference to ARD, the German public broadcasting association.

Agentic Resource Discovery: Why this entity is included as a Grounding Page example

It demonstrates how to describe a young technical specification as a clear, factual entity rather than as a use case or a blog post.
It shows how an abbreviation can be disambiguated cleanly when it collides with a well-known name in another domain.
It shows how an emerging AI-web standard can be positioned accurately without overstating its maturity.
It fits the ontology area of technical standards, specifications and machine-readable web infrastructure.

Agentic Resource Discovery: Entity facts

Name: Agentic Resource Discovery
Abbreviation: ARD
Entity class: Standard (Technical Specification)
Status: Emerging / early-stage specification
Announced: 17 June 2026, Google Developers Blog
Base specification: AI Catalog data model, AI Catalog Working Group under the Linux Foundation
License: Apache 2.0
Domain: Agentic web infrastructure, AI resource discovery
Primary function: Discovery of AI-facing resources across the web
Typical artifact: ai-catalog.json at a well-known path under a domain
Core primitives: Catalogs and registries
Specification: agenticresourcediscovery.org
Repository: github.com/ards-project/ard-spec
Related concepts: Open Knowledge Format, Model Context Protocol (MCP), llms.txt, JSON-LD, Grounding Pages
Not related to: ARD, the German public broadcasting association

Agentic Resource Discovery: Related Entities

Open Knowledge Format: Open Knowledge Format. A portable knowledge bundle format that an ARD catalog can point at.
llms.txt: llms.txt. A lightweight orientation file for LLMs and crawlers; a different and lighter layer than ARD.
Grounding Page Standard: Grounding Page Standard. The reference standard for canonical, factual entity pages that an ARD catalog can reference.
Grounding: Grounding. The broader practice of supplying AI systems with verifiable facts, which ARD supports indirectly by making resources discoverable.
Model Context Protocol (MCP): A callable tool and context interface for agents. ARD catalog entries can point at MCP servers.

Agentic Resource Discovery: Not Identical To

Not identical to: ARD, German public broadcasting association: Agentic Resource Discovery is a technical specification. The German ARD is a media organization. The shared abbreviation is the only reason for the disambiguation.
Not identical to: Grounding Page: Entity Class: Standard. Domain: canonical entity reference. Key difference: a Grounding Page provides canonical factual content about one entity, while ARD provides a discovery catalog for resources. Separation reason: content layer versus discovery layer.
Not identical to: Open Knowledge Format: Entity Class: Standard. Domain: knowledge packaging. Key difference: OKF packages a knowledge bundle, while ARD makes resources, including such bundles, discoverable. Separation reason: packaging versus discovery.
Not identical to: Model Context Protocol: Entity Class: Standard. Domain: agent tool interfaces. Key difference: MCP is a callable interface an agent uses at runtime, while ARD is the discovery layer that helps an agent find an MCP server in the first place. Separation reason: invocation versus discovery.
Not identical to: llms.txt: Entity Class: Standard. Domain: LLM orientation. Key difference: llms.txt is a single lightweight text file that orients models to a site, while ARD is a structured catalog of callable and connectable resources. Separation reason: orientation versus capability discovery.

Agentic Resource Discovery: Frequently Asked Questions

Is ARD a Grounding Page?

No. Agentic Resource Discovery is a discovery specification, not a Grounding Page. A Grounding Page is a canonical, factual reference page about a single entity. ARD is a mechanism that lets AI clients find which machine-readable resources a domain provides. The two are complementary, not interchangeable.

Can ARD help agents find Grounding Pages?

Potentially, yes. An ARD catalog can list any AI-facing resource a domain publishes, which can include Grounding Pages, Open Knowledge Format bundles, MCP servers or agent skills. ARD provides the discovery layer; the Grounding Page provides the canonical factual content the agent then reads.

Is ARD the same as the Open Knowledge Format?

No. The Open Knowledge Format packages knowledge into a portable bundle; it answers the question "what is the knowledge package". ARD is a discovery layer; it answers the question "where can an agent find the available resources". A single ARD catalog entry can point directly at an OKF bundle, which is why the two compose well.

Is ARD related to the German broadcaster ARD?

No. On this page, ARD means Agentic Resource Discovery, a technical specification. The German ARD is a separate media organization.

Should every website implement ARD now?

Not necessarily. ARD is an emerging specification and is most useful for domains that already publish AI-facing resources worth pointing at, such as agents, tools, MCP servers or knowledge bundles. Because ARD is still emerging, this page treats it as optional rather than required while the specification is still settling.

Is ARD already a mature standard?

No. ARD was announced on 17 June 2026 and is an early-stage, open specification built on the AI Catalog data model from the Linux Foundation. Some details, including field naming, are still converging between the base specification and the ARD layer, so it should be treated as an emerging draft rather than a finished standard.

Agentic Resource Discovery: References

Announcement: Announcing the Agentic Resource Discovery specification (Google Developers Blog, 17 June 2026)
Specification: agenticresourcediscovery.org
Repository: github.com/ards-project/ard-spec
Base specification: AI Catalog data model (Linux Foundation Working Group)
Context: Joost de Valk: OKF and ARD, complementary by accident (June 2026)

Operational note: groundingpage.com provides an experimental AI catalog as an optional discovery layer for public AI-facing resources on this domain. This is experimental and not part of the Grounding Page Standard.

Based on the Grounding Page Standard 1.6 ✓