Guides

Agentic Resource Discovery: How Agents Find Resources

A practical guide to finding Agent Skills, MCP servers, tools, and agents with the first public ARD finders.

Published August 2, 2026

On June 17, 2026, Google, GitHub, Microsoft, Hugging Face, GoDaddy, and other participants in the agent ecosystem introduced Agentic Resource Discovery, or ARD.

AI clients already call Skills, MCP servers, APIs, workflows, tools, and other agents. These capabilities live across public repositories, vendor platforms, marketplaces, and private company systems. Finding the right one still requires a reliable place to search.

ARD provides a shared discovery layer. An AI client can describe a task in natural language and search a collection of available capabilities. Results explain what each resource does, who provides it, its type, and where to find it.

The first public implementations are available now. GitHub Agent Finder searches resources connected to GitHub. Hugging Face Discover searches Skills, applications, and MCP servers from the Hugging Face ecosystem.

What does Agentic Resource Discovery cover?

The Agentic Resource Discovery specification calls an external capability an agentic resource. The term covers:

• Agent Skills
• MCP servers
• agents
• APIs
• plugins
• workflows
• tools
• resource catalogs

A larger agent ecosystem changes continuously. Vendors publish new Skills, teams add private workflows, and public repositories introduce specialized agents and MCP servers. ARD lets a client search a collection for the current task. Public finders can cover broad ecosystems, while private finders can limit results to approved company resources.

How to find resources with GitHub Agent Finder

GitHub Agent Finder provides the quickest browser-based way to try ARD.

1. Open the Agent Finder catalog.
2. Describe the task in the search field.
3. Include the job, input, or expected output.
4. Open the most relevant results.
5. Check the resource type, description, publisher, and source repository.
6. Review the resource before connecting or installing it.

Start with a task-shaped query:

create a presentation from a research report

Other useful queries include:

find an MCP server for querying PostgreSQL

find a Skill that reviews Terraform for security problems

find a tool that extracts tables from scanned PDFs

GitHub Copilot can also use Agent Finder from Agent mode. The ARD connector for Copilot provides an explicit command:

/agentfinder find an MCP server for querying PostgreSQL

Copilot presents the matches and leaves installation under user control. Organizations can point Agent Finder at a private registry so employees search an approved internal collection.

How to find resources with Hugging Face Discover

Hugging Face Discover is useful for tasks involving models, datasets, training, inference, AI applications, or MCP servers hosted on Hugging Face.

Install the Hugging Face CLI and Discover extension:

uv tool install -U huggingface_hub
hf extensions install huggingface/hf-discover

Search for a Skill:

hf discover search "fine tune a language model" --kind skill

Search for an MCP server:

hf discover search "generate an image" --kind mcp

Add the domain, input, or output to help the finder distinguish the task. Queries such as train a sentence-transformer on a biomedical dataset and transcribe a multilingual audio file with timestamps carry more intent than category names alone.

How to find Agent Skills with gh skill

gh skill provides a GitHub-specific search and installation flow for Agent Skills. It is available in GitHub CLI 2.90.0 and newer.

Check the installed version:

gh --version

Search public GitHub repositories for Skills:

gh skill search presentation

Narrow the search to a publisher:

gh skill search terraform --owner hashicorp --limit 5

Preview a Skill and its file tree:

gh skill preview getsentry/skills presentation-creator

Install the Skill for Codex in the current project:

gh skill install getsentry/skills presentation-creator --agent codex --scope project

List installed Skills and check for updates:

gh skill list --agent codex --scope project
gh skill update --dry-run

gh skill search focuses on public SKILL.md files stored on GitHub. Agent Finder and Hugging Face Discover can return several kinds of agentic resources. When an ARD result points to a GitHub-hosted Skill, gh skill preview and gh skill install provide the next steps.

How to write better discovery queries

Discovery works better when the query describes the job. Include the action, subject, input, output, or environment that distinguishes the task.

Run the same task through more than one finder when coverage matters. Compare publisher, resource type, source, and the match to the requested workflow.

What to check before installing a result

Discovery produces candidates. Review each candidate before it enters the agent environment.

1. Publisher: confirm the repository owner or product domain.
2. Source: follow the result to its original repository or vendor page.
3. Resource type: identify the Skill, MCP server, agent, API, or workflow.
4. Instructions: read SKILL.md, server cards, and setup documentation.
5. Scripts: inspect commands, downloads, file access, and network destinations.
6. Secrets: identify required API keys and their storage location.
7. Version: prefer a release, tag, or commit that can be pinned.
8. Permissions: limit access to the files, tools, and services required for the task.

For GitHub-hosted Skills, gh skill preview provides a convenient inspection step before installation.

Start with the task, then verify the source

GitHub Agent Finder offers a browser and Copilot search experience. Hugging Face Discover provides a command-line path into machine learning Skills, applications, and MCP servers. gh skill handles discovery and installation for Skills stored on GitHub.

Describe the task, search an approved finder, inspect the matches, verify the source, and connect only the selected resource.