GitHub and Google are among the companies backing a new open specification for finding and verifying tools, skills, and other agents across the web.
The specification, called Agentic Resource Discovery, or ARD, provides a common way to publish, search, and validate AI capabilities across platforms. GitHub said the approach reduces the need to hand-wire MCP servers, skills, canvases, agents, and tools into each workflow.
The specification was developed with participation from several companies, including Microsoft, Google, GoDaddy, Hugging Face, and GitHub. Microsoft said ARD creates a common layer for publishing, indexing, and discovering AI capabilities.
GitHub’s Agent Finder for Copilot implements the ARD specification. GitHub said the feature lets Copilot search an index of available AI resources and return ranked matches from a selected catalogue.
A shared discovery layer for agents
Microsoft described ARD as a response to fragmented registries for AI resources, including MCP servers, skills, tools, and agents.
Microsoft said GitHub’s Agent Finder allows Copilot to discover and call MCP servers, skills, tools, and agents at runtime. It said this helps avoid loading unnecessary resources into the context window.
ARD defines how organisations can publish catalogues of available resources under their own domains. These catalogues can describe agents, tools, skills, and MCP servers. They can also include A2A agents, OpenAPI tools, or nested catalogues.
How ARD works
A provider publishes an ai-catalog.json file at a known location on its domain. That file describes the resources available from the provider and gives agent clients the information needed to discover and connect to those resources.
Registries then crawl and index these catalogues. When an agent needs a capability, it can query a registry using a plain-language request or fetch a catalogue directly from a known domain.
GitHub said developers can describe a task in plain language. Agent Finder then searches an index of available AI resources and returns ranked matches. Copilot can pull in the selected capability from a catalogue chosen by the user or organisation.
The registry returns matching capabilities together with metadata that allows the client agent to verify the publisher before connecting. Google said ARD is designed to support cryptographic verification in production environments, including checks on the publisher’s identity.
After verification, the agent connects directly to the selected resource using its native protocol or API. ARD does not replace those protocols, according to Google. The specification covers discovery and trust metadata before the connection is made.
GitHub brings ARD into Copilot
Agent Finder works against a registry selected by the user or enterprise. GitHub said it can search GitHub’s curated public catalogue or a private registry of internal resources, with results scoped to the chosen source.
GitHub said enterprises can use managed Copilot settings to define which resources agents are allowed to discover and use. Agent Finder only surfaces resources permitted under those settings.
The feature does not automatically install or connect tools. GitHub said Agent Finder surfaces resources, while users decide what gets added to the workflow.
Agent Finder is available across all GitHub Copilot plans. Because it implements ARD, GitHub said any registry or AI client can adopt the same discovery model.
Governance around agent access
GitHub also added enterprise-managed controls for Copilot agents. Administrators can use a disableBypassPermissionsMode setting to stop GitHub Copilot CLI and VS Code from automatically skipping permission prompts. The setting applies to users licensed through Copilot Business or Copilot Enterprise.
The setting can be applied through a private .github-private repository. GitHub said Copilot automatically pulls and applies the configuration.
Google said its Agent Registry in Gemini Enterprise Agent Platform will support searching, discovering, and hosting agentic resources. These resources include agents, skills, MCP servers, and other tools.
Google said the registry will support globally unique namespaced URNs, agentic egress policies, and pinned tools or specifications. It will also use Agent Identity to verify ARD trust manifests. These manifests are used to verify agent authenticity.
Native ARD support for Gemini Enterprise Agent Platform is expected in the coming months. Google said this will allow organisations to connect internal registries to a broader federated network.
Copilot’s agent tooling expands
GitHub made its Copilot SDK generally available earlier in June. The SDK gives developers access to the agent runtime and supports planning, tool invocation, file edits, streaming, and multi-turn sessions.
GitHub also made the Copilot app generally available for macOS, Windows, and Linux, describing it as a desktop environment for agent-driven development built natively on GitHub.
GitHub has also made auto model selection generally available in Copilot Chat on github.com and the GitHub mobile app. The feature chooses a model based on task complexity, model availability, plan access, and administrator policies.
The specification is licensed under Apache 2.0 and builds on the AI Catalog data model. Google credited the AI Catalog Working Group under the Linux Foundation, along with launch partners that contributed to the specification and reference implementations.
(Photo by Yancy Min)
See also: GitHub brings agentic workflows to GitHub Actions
Want to dive deeper into the tools and frameworks shaping modern development? Check out the AI & Big Data Expo, taking place in Amsterdam, California, and London. Explore cutting-edge sessions on machine learning, data pipelines, and next-gen AI applications. The event is part of TechEx and co-located with other leading technology events. Click here for more information.
Developer Tech News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.