AI Agent Discoverability Checker

Discovery comes before everything else

An AI agent cannot use a capability it cannot find. Today most agents work from whatever tools were hardcoded into them at build time; ARD replaces that with search. Registries crawl catalogs published at /.well-known/ard.json, index each entry by its representative queries, and answer an agent asking "what can book a flight?" with the resources that declared they can. If your MCP server, A2A agent or skill is not in a catalog, you are relying on someone wiring you in by hand.

The scoring here mirrors that pipeline: presence at the current path, a valid manifest, entries carrying the terms the spec requires, the query signals the index is built from, recognized resource types (MCP, A2A, skills), the secondary discovery mechanisms (link tag, Agentmap, in-page JSON-LD), a trust manifest, and plain HTTPS. Each row names the fix when it fails - most take minutes, starting with generating a catalog.

The legacy trap

ARD v0.9 publishers did everything right and still ended up half-invisible: v0.91 moved the manifest from ai-catalog.json to ard.json, and consumers are only required to look at the new path. If this checker finds you on the legacy path only, serve the identical file at both locations while the ecosystem transitions. It is a five-minute change that doubles your chances of being found.

How to use this tool

1. Enter any website URL and click Check discoverability.
2. Read the score and the per-check table - every failed row names its fix.
3. Browse the capabilities list: that is exactly what a registry would index from the catalog it found.
4. No catalog yet? Generate one with the ARD Catalog Generator, publish it, and re-check.
5. Copy the report for your team or your records.

Related Tools