Google and industry partners have announced the Agentic Resource Discovery (ARD) Specification, an open standard designed to enable AI agents to publish, discover, and verify external tools, APIs, and services across organizational boundaries. The specification addresses a growing gap in AI agent infrastructure, where capabilities are widely distributed but lack a common discovery and trust mechanism.
While protocols such as the Model Context Protocol (MCP) define how an AI agent invokes a tool, ARD addresses an earlier stage in the lifecycle: how agents discover those tools in the first place. Rather than replacing existing standards, ARD is designed as a complementary discovery layer that works across frameworks and providers.
Srinivas Krishnan, Distinguished Engineer at Google Cloud, highlights the motivation for ARD:
The problem is simple to state and hard to solve, especially in the enterprise, where the answer can't just be ‘find something that works.’ It has to be governed, with security and identity built in rather than bolted on.
The specification introduces two core constructs: catalogs and registries. Organizations publish a machine-readable ai-catalog.json file within their domain that describes available capabilities such as tools, APIs, skills, and agent endpoints. Registries aggregate these catalogs and enable agents to search based on task intent instead of relying on hardcoded integrations or static endpoint lists. This allows agents to locate relevant resources across organizational boundaries while remaining compatible with execution standards such as MCP and OpenAPI.
How ARD Works (Source: Google Blog Post)
Trust and verification are central to the design. ARD includes domain-based ownership and verification mechanisms so agents can validate the authenticity of discovered resources before establishing connections. This is intended to reduce risk in environments where autonomous agents may trigger actions across third-party services and enterprise systems.
From Reddit community discussions, one perspective highlights the value of standardization:
A uniform baseline protocol makes it easier to build alternatives without needing to interpret many different proprietary documentation formats.
However, discussions also point out that the effectiveness of such systems will depend on the quality of exposed tools and the access or pricing models associated with them.
The specification was developed with contributions from Microsoft, GitHub, Hugging Face, Cisco, Databricks, GoDaddy, NVIDIA, Salesforce, ServiceNow, and Snowflake. Early implementations have already emerged, including GitHub’s Agent Finder in Copilot and Hugging Face’s Discover Tool, both leveraging ARD for runtime capability discovery.
Jennifer Marsman, Principal Engineer in AI at Microsoft, notes:
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.
The ARD specification is currently available with reference implementations and documentation, allowing organizations to experiment with publishing capability catalogs and exploring the federation model defined in the specification. It includes schemas, trust mechanisms, and guidelines for interoperability across discovery services. The broader ecosystem is expected to evolve through community contributions, including implementation feedback and extensions to the schema and governance model.