ARD / ai-catalog.json Adoption Tracker
In short: This is a living, independently-measured tracker of who actually publishes an Agentic Resource Discovery catalog. We probe a sample of major sites for the three ARD discovery surfaces (
/.well-known/ai-catalog.json, arobots.txtAgentmapdirective, and a<link rel="ai-catalog">tag) and publish the per-company results with a date stamp. Latest snapshot, 2026-06-18: 0 of 39 sites — including all eleven named ARD working-group members — publish a discoverable catalog. We re-run it on a cadence, expand the sample, and record every implementer here the moment one appears.
This is a Reference page: an independent verification log, not a play. Tracking who adopts a standard is its own kind of evidence, separate from the playbook and the pilot experiments in the Data Room. We are spinning adoption tracking out into its own dated artifact so the numbers stay honest and quotable as the landscape moves. ARD is the first standard we track this way; more will follow.
The current snapshot (2026-06-18)
| Discovery surface | Adoption | Detail |
|---|---|---|
/.well-known/ai-catalog.json (valid catalog) | 0 / 39 | No site served parseable JSON with the required specVersion + entries shape |
robots.txt Agentmap: directive | 0 / 39 | Not one site, including Google's |
<link rel="ai-catalog"> in homepage <head> | 0 / 39 | None |
Spec announced ~2026-05-28; first measured ~3 weeks later. A 200 is not adoption: several sites return a 200 from an SPA catch-all, or 401/403 from a CDN edge, at the well-known path. We parse the body and require the ARD shape before counting it, so none of those qualify.
Who has implemented ai-catalog.json?
Nobody in our sample, yet. This is the table that matters; it stays empty until a real catalog appears, and then it fills with names and dates.
| Company | Domain | Valid catalog | First seen | Entries | Notes |
|---|---|---|---|---|---|
| (none yet) | — | — | — | — | Be the first. See how to get listed. |
When a site ships one, this row becomes a dated entry: the company, the date we first verified it, how many resources it lists, and what types (MCP server, A2A agent, API). That is the leaderboard this page is built to grow.
Full sample results
The sample is deliberately friendly to ARD: every company named in the working group, plus the developer-tool and infrastructure companies that lead every other agent-readiness curve in this guide. If anyone had shipped a catalog, this sample would catch it. Legend: ✗ = absent, ✓ = present.
ARD working-group members (13 domains)
| Domain | ai-catalog.json | Agentmap | <link> |
|---|---|---|---|
| google.com | ✗ | ✗ | ✗ |
| cloud.google.com | ✗ | ✗ | ✗ |
| microsoft.com | ✗ | ✗ | ✗ |
| azure.microsoft.com | ✗ | ✗ | ✗ |
| github.com | ✗ | ✗ | ✗ |
| huggingface.co | ✗ | ✗ | ✗ |
| nvidia.com | ✗ | ✗ | ✗ |
| salesforce.com | ✗ | ✗ | ✗ |
| servicenow.com | ✗ | ✗ | ✗ |
| snowflake.com | ✗ | ✗ | ✗ |
| databricks.com | ✗ | ✗ | ✗ |
| cisco.com | ✗ | ✗ | ✗ |
| godaddy.com | ✗ | ✗ | ✗ |
That every author of the spec is a ✗ three weeks after publishing it is the single most important data point on this page. Specs the sponsors haven't shipped themselves are a coin-flip to matter; treat that as the honest prior.
Developer-tool & infrastructure companies (26 domains)
| Domain | ai-catalog.json | Agentmap | <link> |
|---|---|---|---|
| anthropic.com | ✗ | ✗ | ✗ |
| openai.com | ✗ | ✗ | ✗ |
| stripe.com | ✗ | ✗ | ✗ |
| vercel.com | ✗ | ✗ | ✗ |
| nextjs.org | ✗ | ✗ | ✗ |
| supabase.com | ✗ | ✗ | ✗ |
| prisma.io | ✗ | ✗ | ✗ |
| clerk.com | ✗ | ✗ | ✗ |
| better-auth.com | ✗ | ✗ | ✗ |
| resend.com | ✗ | ✗ | ✗ |
| convex.dev | ✗ | ✗ | ✗ |
| orm.drizzle.team | ✗ | ✗ | ✗ |
| tailwindcss.com | ✗ | ✗ | ✗ |
| hono.dev | ✗ | ✗ | ✗ |
| upstash.com | ✗ | ✗ | ✗ |
| dodopayments.com | ✗ | ✗ | ✗ |
| bun.sh | ✗ | ✗ | ✗ |
| modelcontextprotocol.io | ✗ | ✗ | ✗ |
| context7.com | ✗ | ✗ | ✗ |
| smithery.ai | ✗ | ✗ | ✗ |
| langchain.com | ✗ | ✗ | ✗ |
| cloudflare.com | ✗ | ✗ | ✗ |
| shopify.com | ✗ | ✗ | ✗ |
| twilio.com | ✗ | ✗ | ✗ |
| auth0.com | ✗ | ✗ | ✗ |
| mintlify.com | ✗ | ✗ | ✗ |
Methodology
For each domain, on the snapshot date, we run three checks:
GET https://{domain}/.well-known/ai-catalog.json, following redirects. We record the final HTTP status and content type, then parse the body and require the ARD shape (specVersionpresent andentriesan array) before counting it. This rejects SPA catch-alls that return200with HTML, and CDN401/403pages.GET /robots.txt, scanned for anAgentmap:directive (the ARD analogue ofSitemap:) or anyai-catalogreference.GET /, scanned for<link rel="ai-catalog">in the document head.
A domain counts as an implementer only if check 1 returns a valid catalog. The other two are tracked because the spec lists them as optional discovery aids.
Biases, stated plainly. The sample skews to US developer-tool and big-tech domains; it is not a random sample of the web. We probe apex domains and a few documented subdomains (cloud.google.com, azure.microsoft.com), so a catalog hosted on an undocumented subdomain or docs host could be missed. Shares are point-in-time; CDN responses vary between runs (we saw the same path return 403 one day and 200 the next, neither a catalog). We publish the script so anyone can reproduce or challenge a number.
Snapshot history
Each run is appended here so the trend is visible over time. Newest first.
| Date | Sample size | Valid catalogs | Agentmap | <link> | Notes |
|---|---|---|---|---|---|
| 2026-06-18 | 39 | 0 | 0 | 0 | First census. ~3 weeks after the spec announcement. All 11 working-group members absent. |
The next runs will widen the sample (toward several hundred domains across more categories) and add columns as implementers appear. The cadence is at least quarterly, and immediate when a major vendor announces support.
Reproduce it yourself
The probe is a single dependency-free script. Run it against the default sample, or extend it:
# default 39-domain sample
node apps/synscribe/scripts/ard-adoption-census.mjs
# add your own domains (comma-separated)
node apps/synscribe/scripts/ard-adoption-census.mjs yourdomain.com,partner.com
It prints a per-domain table and the headline counts. The "valid catalog" check is an AND on specVersion + an entries array, so a stray 200 or partial JSON never inflates the number.
How to get on this list
We want this table to fill up. Two ways an entry appears:
- Ship a valid catalog. Follow the Play 12 adoption playbook: host a conformant
/.well-known/ai-catalog.json, anchor your URN to your domain, and engineer yourrepresentativeQueries. We'll catch it on the next run. - Tell us. If you've shipped one (or spotted one in the wild we missed), let us know and we'll verify and date it. Synscribe publishes its own catalog at
/.well-known/ai-catalog.json— we practice what we track.
FAQ
What is the current ai-catalog.json adoption rate?
As of 2026-06-18, 0 of 39 surveyed top sites publish a discoverable ai-catalog.json, an Agentmap robots directive, or a <link rel="ai-catalog"> tag — including all eleven named ARD working-group members. This is our first dated snapshot; we re-run it on at least a quarterly cadence and expand the sample over time.
Has anyone implemented ARD yet? Not in our sample. The implementers table on this page is empty and will be filled, with names and dates, the moment a valid catalog appears. If you've shipped one, tell us and we'll verify it.
How do you decide a site "has" a catalog?
We fetch /.well-known/ai-catalog.json, parse the body, and require the ARD shape: a specVersion field and an entries array. A 200 response alone doesn't count, because SPA catch-alls and CDN error pages return 200/403 with non-catalog bodies. The check is reproducible via our open script.
Why track adoption separately from the playbook? Because independent, dated verification is its own evidence. Whether a standard gets adopted is a different question from how to use it, and it deserves a clean, quotable record that updates as reality changes. This page is the first of our adoption trackers; the Data Room holds the rest of our numbers.
Last verified 2026-06-18. This is a living tracker; we re-run the census on a cadence and date every snapshot. Methodology and caveats above; reproduce with the open script.
Part of The Complete Playbook to Agentic Discovery.
Stay ahead of the agents. We re-test this playbook quarterly and publish what changed: new data, busted myths, ranking shifts. Get the update digest →
Want this done for you? Synscribe runs agentic-discovery programs for B2B SaaS and developer platforms. Talk to us →