Google and Microsoft Lead 11 Tech Companies in Launching Agentic Resource Discovery Specification

Google, Microsoft, and a coalition of 11 tech companies published Agentic Resource Discovery (ARD) on June 17, an open specification that gives AI agents a standard way to find tools, skills, and other agents at runtime instead of relying on pre-wired integrations.

ARD sits in front of existing protocols like MCP (how agents call tools) and A2A (how agents talk to each other). It solves a problem that has kept agentic AI from scaling: agents can only use tools that developers manually connected ahead of time.

Microsoft's Ramanathan Guha described the bottleneck bluntly on the Command Line blog: "AI can only use what it's been explicitly wired to use. Everything else may as well not even exist."

The specification defines two primitives. Organizations publish an ai-catalog.json file on their own domain listing available capabilities.

Registries then crawl those catalogs and expose a search API that agents query in plain language. Domain ownership doubles as cryptographic identity, letting agents verify a publisher before connecting.

After verification, ARD steps aside and the agent connects directly using the tool's native protocol.

Eleven companies co-developed the spec, including Cisco, Databricks, GitHub, GoDaddy, Google, Hugging Face, Microsoft, NVIDIA, Salesforce, ServiceNow, and Snowflake. ARD is licensed under Apache 2.0 and builds on the AI Catalog data model maintained by a working group under the Linux Foundation.

GitHub shipped a reference implementation the same day. Agent Finder for Copilot lets developers describe a task in natural language and returns ranked matches from a curated public catalog or a private enterprise registry.

Hugging Face released a Discover Tool that provides semantic search across thousands of skills and MCP servers on its Hub. Google's Agent Registry within the Gemini Enterprise Agent Platform is set to add native ARD support in the coming months.

Microsoft positioned ARD as the missing layer between fragmented registries that have kept AI resources siloed. "We're leading with an open specification because we want to encourage open ecosystems," the company wrote.

"Not walled catalogs. Not curated directories that can't keep up." The move mirrors the early web's transition from hand-curated directories to search engines.

ARD applies that model to agentic resources, where the ecosystem of available MCP servers, APIs, and agents already numbers in the hundreds of thousands. The specification is a v0.9 draft, with contributors inviting feedback through the project's GitHub repository.