Design.dev design.dev
AI Tools

ARD Catalog GeneratorExperimental draft spec

Create an Agentic Resource Discovery catalog that makes your AI resources discoverable to agents and registries. Generate a ready-to-publish ai-catalog.json for /.well-known/ai-catalog.json.

Experimental. ARD is an emerging draft specification. This generator produces a discovery manifest — not an agent rule file or execution config. Validate the generated file against the current spec before using it in production.

1 Host information The catalog operator

Accepts acme.com, https://acme.com, or www.acme.com/docs — normalized to a clean domain for URN identifiers like urn:ai:acme.com:server:weather.

2 Catalog entries The resources you publish

Quick-start templates
Spec Validator
Valid
No issues — your catalog looks spec-compliant.
ai-catalog.json 

							

						

					


					

						
					


					
					

						
Where to publish ai-catalog.json

						

							
your-site/
└── .well-known/
    └── ai-catalog.json   ← served at /.well-known/ai-catalog.json

						

						

							
Publishing checklist

							
    
								
  • 
									Save the file as
									ai-catalog.json
								
    • 
								
  • 
									Upload it to the well-known path
									https://<domain>/.well-known/ai-catalog.json
								
    • 
								
  • 
									Serve with the correct content type
									Content-Type: application/json
								
    • 
								
  • 
									Serve over HTTPS and allow cross-origin reads
									Access-Control-Allow-Origin: *
								
    • 
								
  • 
									Optional DNS TXT fallback
									_catalog._agents.<domain> TXT "url=https://example.com/path/ai-catalog.json"
								
    • 
								
  • 
									Optional dynamic search SRV
									_search._agents.<domain> SRV 443 search.<domain>
								
    • 
							

						

					


					

						
							
The resource

							
MCP Config Generator →

							
ARD makes MCP servers discoverable. This builds the MCP config the resource actually runs on.

						
						
							
The discovery layer below

							
llms.txt Generator →

							
llms.txt points LLMs at your pages. ai-catalog.json points agents at your callable resources.

						
						
							
Skill artifacts

							
SKILL.md Generator →

							
Author the Skills you reference as application/ai-skill+md entries in your catalog.

						
						
							
For repos

							
AGENTS.md Generator →

							
ARD sits above per-repo configs as a cross-org discovery manifest for your resources.

						
					

				

			

		


		

			

				

					
Frequently asked questions

					
Quick answers about ai-catalog.json, the ARD spec, and how it relates to MCP and llms.txt.

				

				

					

						What is ai-catalog.json?
						
ai-catalog.json is the manifest file used by Agentic Resource Discovery (ARD) to list the AI resources available from a domain. It gives agents and registries a structured way to find your MCP servers, A2A agents, Skills, APIs, nested catalogs, and registries. Use it when you publish AI-callable resources and want them discoverable outside a single app, IDE, or agent framework — it sits above MCP configs, AGENTS.md, SKILL.md, and llms.txt as a discovery layer.

					

					

						What should each entry include?
						
A stable domain-anchored URN identifier, a human-readable display name, and a media type describing the artifact, plus exactly one of a URL to the artifact or inline JSON data. Recommended additions: a clear description, capabilities and tags for filtering, 2–5 representative natural-language queries for semantic discovery, and optional trust metadata for identity, provenance, and attestations.

					

					

						Is ai-catalog.json the same as llms.txt?
						
No. llms.txt helps language models understand important pages on a site. ai-catalog.json lists agentic resources such as tools, agents, Skills, APIs, catalogs, and registries.

					

					

						Does ARD replace MCP?
						
No. ARD helps agents discover MCP servers and other resources. The resource still runs through its own native protocol.

					

					

						Does ARD make my tool safe or verified?
						
No. ARD can carry trust and provenance metadata, but the registry or client still decides what to trust.

					

					

						Where should I host the file?
						
The standard location is https://<your-domain>/.well-known/ai-catalog.json.

					

					

						What is a representative query?
						
A short natural-language example of what a user might ask when they need this resource. These examples help discovery systems rank resources semantically.

					

				

			

		

	


	

		

			
		

	


	

	
Copied to clipboard!