Key Points
- Google published the Agentic Resource Discovery (ARD) specification on June 17, 2026.
- ARD lets AI agents find and verify tools, APIs, MCP servers, and other agents at runtime.
- It defines two primitives: a static ai-catalog.json manifest and a crawling registry API.
- Backers include Microsoft, Nvidia, GitHub, Hugging Face, Salesforce, Snowflake, and Cisco.
The AI industry spent the last two years racing to build agents. It is now trying to agree on how those agents find each other. On June 17, Google published the Agentic Resource Discovery specification — ARD — an open standard for publishing, discovering, and verifying AI capabilities across the web, with a launch roster that reads like a who's-who of the cloud and AI business: Microsoft, Nvidia, GitHub, Hugging Face, Salesforce, Snowflake, Cisco, Databricks, GoDaddy, and ServiceNow.
The problem ARD is trying to solve
As agents proliferate, they face three questions they currently can't answer on their own: where does the right capability live, which one should they use, and how do they verify it is safe to connect to. Today those answers live in fragmented, vendor-specific registries. ARD's pitch, per the specification site, is "a standard way for agents to discover capabilities across organizational boundaries and establish trust."
It does that with two primitives. First, a static ai-catalog.json manifest that an organization hosts at a well-known path on its own domain, listing the MCP servers, agents, and APIs it offers — with domain ownership serving as the cryptographic identity. Second, a registry API that crawls and indexes those published catalogs and returns ranked matches to natural-language queries, effectively a search engine for the agentic web. The model is federated: anyone can run a registry, and registries can cross-reference each other, with no central gatekeeper. The spec is licensed Apache 2.0 and builds on the AI Catalog data model from the Linux Foundation's AI Catalog Working Group.
Why a shared standard is the news
The read: the interesting part isn't the JSON — it's the guest list. ARD ships with client guides for Claude, ChatGPT, GitHub Copilot, Microsoft Copilot, and Gemini, which means the same labs that compete on models are quietly agreeing to share the plumbing underneath them. The agent economy needs something like DNS for capabilities, and whoever authors that standard shapes how discovery and trust work for everyone downstream. That those rivals all signed on is the same competitive logic visible elsewhere in the stack: OpenAI is racing to own its silicon with the Broadcom-built Jalapeño chip and recently restructured its Microsoft partnership to end cloud exclusivity, while Anthropic has confidentially filed to go public. Everyone wants to own a layer; ARD is a bet that discovery should be the shared one.
Every resource an agent discovers is also an identity someone has to govern, which is the market Cyera moved on in July with a $1 billion letter of intent for Oasis Security.
Source: Google Developers Blog, AgenticResourceDiscovery.org, and Hugging Face.
Comments