For makers and artists, the Agentic Resource Discovery (ARD) specification marks the end of the “install-first” era for AI tools. It shifts the paradigm from manually configuring static catalogs to dynamic, intent-based search. Instead of hardcoding server URLs or relying on limited context windows to sift through tool descriptions, agents can now query a federated registry at runtime. This allows your AI workflows to discover and utilise capabilities—from custom skills to MCP servers—across the wider ecosystem without pre-configuration. The standard is not a marketplace; it is an open protocol developed by contributors including Microsoft, Google, GoDaddy, and Hugging Face, enabling any tool or agent to participate.
Solving the discovery bottleneck
The current approach to agent capabilities relies on static installation. A developer typically hardcodes a Model Context Protocol (MCP) server URL into a configuration file, or a user connects a service via a plugin. While functional for a daily handful of tools, this model fails to scale when an agent needs to access thousands of ad-hoc services. The alternative—dumping every available tool description into the Large Language Model’s context window—is constrained by context budgets and often results in poor disambiguation.
ARD moves the selection logic outside the LLM. A registry indexes capabilities using richer signals, such as publisher identity, representative queries, and compliance attestations, exposing a REST endpoint for natural language search. The model simply invokes what the search returns. This transition allows an agent to dynamically find the correct capability, reaching a growing ecosystem of MCP tools and A2A agents without the need for individual pre-configuration.
The specification defines two core components:
- A static manifest format,
ai-catalog.json, which lets publishers host their capabilities at a well-known URL. - A dynamic registry API at
POST /searchthat provides live, ranked discovery.
ARD on the Hugging Face Hub
The Hugging Face Discover Tool serves as the reference implementation for ARD. It grants search access to thousands of Skills, ML applications, and MCP Servers, both on the Hub and across other ARD discovery services.
The system combines the Hub’s existing semantic search over Spaces with its Agent Skills, serving results as ARD catalog entries. The Hub already hosts a catalog of Spaces running Gradio apps, MCP servers, and demos. Its semantic search supports an agents=true flag, returning Spaces ranked by agent-oriented metadata, which Discover then translates into the ARD specification.
The adapter applies two primary filters. First, the response includes only Spaces where the runtime stage is RUNNING. Second, the response media type is driven by the request, supporting three formats:
application/ai-skill: The default type. A generatedSKILL.mdwrapping the Space’sagents.md.application/mcp-server+json: An MCP server catalog entry for Spaces taggedmcp-server.application/vnd.huggingface.space+json: Raw Space metadata for clients wishing to handle it themselves.
The skill type involves an additional transformation. Many Spaces ship an agents.md file describing agent interaction. Discover reads this file and wraps it with the frontmatter a skill consumer expects: name, description, and source metadata covering the Space ID, Hub URL, app URL, and the original agents.md link. The result is a skill any skill-aware client can install or load through its normal flow.
For MCP-tagged Spaces, the adapter generates a catalog entry pointing at the Space’s Gradio MCP endpoint over HTTP transport. The URL uses the Space’s runtime domain when the Hub provides one, otherwise the standard .hf.space slug convention.
Using the tool
The discover command is built into the Hugging Face CLI (hf). To get started and grant your agent access:
# Install the Hugging Face CLI tool:
uv tool install huggingface_hub
# Search for resources to train a model
hf discover search "Fine tune a language model"
# Find MCP Servers to generate an image
hf discover search "Generate an image" --json --kind mcp
# Search other registries
hf discover search "Purchase aeroplane tickets" --registry-url <catalog-url>REST API and MCP Tool
You can also search the catalog directly via the REST API or an MCP Server.
The Hugging Face catalog is published at its well-known URL:
https://huggingface.co/.well-known/ai-catalog.jsonTo call the search endpoint directly:
POST https://huggingface-hf-discover.hf.space/searchcurl -s https://huggingface-hf-discover.hf.space/search \
-H "Content-Type: application/json" \
-d '{
"query": {
"text": "fine tune a sentence transformer",
"filter": {
"type": ["application/ai-skill"]
}
},
"pageSize": 5
}'Search for MCP servers:
curl -s https://huggingface-hf-discover.hf.space/search \
-H "Content-Type: application/json" \
-d '{
"query": {
"text": "transcribe some audio",
"filter": {
"type": ["application/mcp-server-card+json"]
}
},
"pageSize": 5
}'Alternatively, connect any MCP Client to search via the MCP endpoint at https://huggingface-hf-discover.hf.space/mcp.
Implications for the specification
ARD separates discovery from execution. The static manifest format is driven by media type, allowing any artifact protocol to ride the same envelope without specification-level changes. The registry API is plain HTTP REST, enabling any client to federate against it. Discover is one of several reference implementations across the ecosystem; because federation is built into the protocol, a search through one service can surface capabilities hosted by another.
The Discover Tool is a working test of this design. It does not invent a new artifact format. It wraps an existing search backend, the Hub, in the specification’s envelope, letting the same Spaces surface as skills or MCP servers depending on the client’s request.
Next steps involve tighter integration with the specification’s federation modes (auto, referrals, none) and Hub-side support for static ai-catalog.json manifests on user and organization profiles. Once implemented, any Space publisher will be able to advertise their capabilities through the standard well-known URI mechanism.
Key takeaways
- ARD replaces static, manual tool installation with a dynamic, intent-based search layer that allows agents to discover capabilities at runtime.
- The Hugging Face Discover Tool acts as a reference implementation, translating the Hub’s semantic search into the ARD standard for Skills and MCP servers.
- The specification is open and federated, meaning any publisher can implement the
ai-catalog.jsonmanifest to join the ecosystem without creating a new marketplace.
Stay ahead of AI. Get the most important stories delivered to your inbox — no spam, no noise.
