# Agentic Resource Discovery — HF, Microsoft, Google open spec for tool lookup

> Agentic Resource Discovery is an open spec from HF, Microsoft, Google and GoDaddy that lets agents find MCP servers, skills, and A2A endpoints at runtime.

- Org: Hugging Face
- Date: 2026-06-17
- Category: ecosystem, tool
- Source: https://huggingface.co/blog/agentic-resource-discovery-launch
- Canonical page: https://ai-tldr.dev/releases/hf-agentic-resource-discovery-spec/

## In short

Hugging Face, Microsoft, Google, and GoDaddy launched Agentic Resource Discovery, an open spec that lets agents find MCP servers, skills, and A2A endpoints at runtime. ARD defines an ai-catalog.json manifest and a POST /search registry API, with a reference Discover Tool on HF.

## What it is

Agentic Resource Discovery (ARD) is an open specification, launched as a draft, that defines how agents discover the tools and skills available to them at runtime — instead of being shipped with a hard-coded list of MCP servers or installed plugins. Hugging Face, Microsoft, Google, and GoDaddy co-developed ARD, and Hugging Face shipped the first reference implementation.

## How it works

ARD has two layers. The first is a Static Manifest Format: a publisher hosts an ai-catalog.json file at a well-known URL (for example huggingface.co/.well-known/ai-catalog.json) that lists the MCP servers, skills, ML apps, and A2A endpoints they expose. The second is a Dynamic Registry API — a POST /search endpoint that takes an intent and returns ranked, federated results across many catalogs. ARD sits above MCP rather than replacing it: MCP defines how an agent CALLS a tool, ARD defines how an agent FINDS the tool first. Hugging Face's hf-discover demo serves the HF catalog through huggingface-hf-discover.hf.space.

## Why it matters

Right now every coding agent ships with a static MCP server list, and adding a new tool means a config edit. ARD reframes tool access as a search problem — an agent gets a fresh intent ('fine-tune a small image model'), queries an ARD registry, and gets back the relevant MCP server, skill, or A2A endpoint without anyone editing settings. With HF, Microsoft, and Google co-signing, ARD is the first credible attempt to make agent capabilities composable across vendors.

## Who it's for

Agent platform builders, MCP server maintainers, anyone publishing AI tools

## Try it

https://huggingface.co/.well-known/ai-catalog.json

## Quick facts

- Makers: Hugging Face, Microsoft, Google, GoDaddy (multi-org)
- Static format: ai-catalog.json at /.well-known/ai-catalog.json
- Dynamic API: POST /search returning ranked discovery results
- Reference implementation: huggingface/hf-discover (open source)
- HF Discover endpoint: huggingface-hf-discover.hf.space
- Relationship to MCP: Layer above — finds the tools MCP then calls

## FAQ

### What is Agentic Resource Discovery?

Agentic Resource Discovery (ARD) is an open draft specification, launched June 17, 2026 by Hugging Face, Microsoft, Google, and GoDaddy, that lets AI agents find available tools, skills, MCP servers, and A2A endpoints at runtime. ARD defines a static ai-catalog.json manifest hosted at a well-known URL plus a dynamic POST /search registry API for federated, ranked discovery.

### How does ARD relate to MCP?

Agentic Resource Discovery complements MCP rather than competing with it. MCP defines how an agent calls a tool once it knows about it; ARD defines how the agent finds the tool in the first place. An agent uses ARD's POST /search to discover candidate MCP servers for an intent, then connects to those servers through the standard MCP protocol.

### Who can publish an ARD catalog?

Any organization can publish an Agentic Resource Discovery catalog by hosting an ai-catalog.json file at /.well-known/ai-catalog.json on their domain. Hugging Face has published the first catalog at huggingface.co/.well-known/ai-catalog.json, exposing semantic search over Skills, ML apps, and MCP servers via the open-source huggingface/hf-discover reference implementation.

## Links

- [Announcement](https://huggingface.co/blog/agentic-resource-discovery-launch)
- [Reference Implementation](https://github.com/huggingface/hf-discover)