Google and partners introduce ARD specification for unified AI agent resource discovery

ARD (Agentic Resource Discovery) is a new open specification co-created by Google and multiple industry partners including Microsoft, GitHub, Hugging Face, Cisco, Databricks, GoDaddy, NVIDIA, Salesforce, ServiceNow, and Snowflake. It addresses a critical gap in AI agent infrastructure: while tools and capabilities are widely available, there is no unified discovery mechanism and trust system. ARD focuses on how agents discover tools earlier in the lifecycle, complementing protocols like MCP that define how to invoke them. The specification introduces two core concepts: Catalog (a machine-readable ai-catalog.json file describing available capabilities) and Registry (aggregating catalogs and allowing agents to search based on task intent). Trust and verification are central, with domain-based resource attribution and verification to reduce security risks. Early implementations include GitHub's Agent Finder in Copilot and Hugging Face's Discover Tool. ARD is not intended to replace existing standards but to work alongside them. Google Cloud engineer Srinivas Krishnan noted the difficulty of solving discovery with proper governance in enterprise environments. Microsoft AI engineer Jennifer Marsman stated ARD does not aim for a single global directory but enables multiple discovery services. The specification is now published with reference implementations and documentation.