source: hugging face blog: agentic resource discovery: let agents search
level: technical
the agentic resource discovery specification is a draft open standard for cataloging and searching agent capabilities across federated registries. it defines a static manifest format called ai-catalog.json and a rest api for live discovery. the goal is to let agents find tools, skills, and other agents at runtime using natural language queries, rather than relying on hardcoded configurations or dumping all tool descriptions into a model's context window.
hugging face has built a reference implementation called the discover tool. it indexes thousands of skills, ml applications, and mcp servers from the hub and other registries. the tool uses the hub's existing semantic search and returns results as ard catalog entries. it supports three media types: skills, mcp server cards, and raw space metadata. for skills, it wraps a space's agents.md file with the frontmatter a skill consumer expects.
users can access discover through the hugging face cli, a rest api, or an mcp server. the cli command `hf discover search` accepts a query and optional filters. the rest api is a post endpoint that returns ranked results. the mcp server allows any mcp client to search the catalog. the specification separates discovery from execution, so any artifact protocol can work within the same framework. future plans include tighter federation and support for static manifests on user profiles.
why it matters: this lets ai agents find and use new tools on the fly, making them more flexible and reducing manual setup for developers.
source: hugging face blog: agentic resource discovery: let agents search