Agentic AI
Google and Industry Partners Launch Standard for AI Agents to Discover and Share Tools
A new open specification enables autonomous AI agents to locate and verify external capabilities across organizational boundaries without hardcoded integrations.
InfoQ
Google and its partners have unveiled the Agentic Resource Discovery (ARD) Specification, an open standard that allows AI agents to find, publish, and trust external tools, APIs, and services beyond their own organization. The initiative tackles a critical infrastructure gap: while AI agent capabilities are scattered across many systems, there's been no unified way for agents to discover and securely access them.
How ARD Fits into the AI Agent Ecosystem#
Existing protocols like the Model Context Protocol (MCP) focus on how agents invoke tools once they know about them. ARD operates at an earlier stage—the discovery phase—helping agents figure out which tools and services are even available in the first place. Rather than displacing other standards, ARD works alongside them as a complementary discovery layer that remains agnostic to specific frameworks or vendors.
The specification centers on two key ideas: catalogs and registries. Each organization creates a machine-readable ai-catalog.json file that lists its available capabilities—tools, APIs, skills, and agent endpoints. Registries then collect these catalogs from across multiple organizations, enabling agents to search by task intent rather than relying on static, hardcoded endpoint lists or manual integration work.
Trust and Verification Built In#
Security and validation sit at the core of the design. ARD incorporates domain-based ownership checks and verification mechanisms that allow agents to confirm a resource's authenticity before attempting to use it. This approach is especially important in enterprise settings where autonomous agents might trigger actions across multiple third-party systems, reducing the risk of rogue or compromised resources.
Microsoft's Jennifer Marsman, Principal Engineer in AI, noted:
"The goal isn't a single global catalog of every resource. There will be many discovery services, each defined by what it indexes, whom it serves, and how it ranks… ARD helps AI clients discover capabilities, but it doesn't replace authentication, authorization, governance, or organizational trust decisions."
