Are merchants publishing A2A or MCP discovery documents?
Frozen merchant panel; dated observations through 2026-08-07.
We confirmed 0 genuine A2A or MCP discovery documents across a 276-merchant panel over 55 sealed collection dates from 2026-06-11 through 2026-08-07. The observed zero is not a count of HTTP responses: every retained 2xx body on the four candidate paths was parsed and had to match a protocol-document shape before it could count.
This census distinguishes genuine protocol documents from HTML soft-200 pages. That distinction is the product: a storefront can return a branded HTML page with status 200 at any unknown path, and a status-only census would report that as adoption.
Coverage receipt. The panel contains 276 selected US-centric consumer-commerce domains, not every merchant. It was observed on 55 of 58 calendar days in the window; missing dates: 2026-06-19, 2026-07-04, 2026-07-15. The newest run sealed 1,932 rows and recorded 144 fetch or robots-gate errors. An unreachable or robots-disallowed path is unobserved, never converted to an absence.
Why raw 200 counts are misleading
| Protocol paths | 2xx on newest seal | 2xx across history | qualifying documents |
|---|---|---|---|
A2A agent cards/.well-known/agent-card.json and /.well-known/agent.json | 46 | 2,595 | 0 |
MCP discovery/.well-known/mcp and /.well-known/mcp/server-card.json | 47 | 2,627 | 0 |
A qualifying A2A document must be a JSON object with a non-empty name, a capabilities object and a skills list. A qualifying MCP document must expose a server collection or a server-info and capabilities pair. An unrelated JSON error does not count. A referenced body that cannot be read makes this page unavailable; the reader never turns missing evidence into zero.
What this record can and cannot establish
The evidence answers one narrow question: what the selected domains served at four public machine-discovery paths when our robots-compliant collector checked. It does not prove that a merchant supports agentic checkout, has no private agent integration, or has no implementation roadmap. OpenAI and Stripe ACP adoption is structurally unobservable from these merchant paths and is outside the count.
The collector keeps bodies privately so the classification can be reproduced, but this page publishes no archived body or excerpt. That boundary prevents role addresses, phone numbers, authored llms.txt text and our own reflected collector contact string from leaking into a future template.
Monthly agentic-surface adoption census
$249/month
For a competitive-intelligence or product-marketing lead at an agentic-commerce infrastructure vendor, the monthly issue covers the frozen merchant panel across A2A, MCP, UCP and llms.txt. It includes per-merchant surface status, first-seen dates, retained content hashes, and confirmed additions or removals.
- CSV plus a dated, hash-sealed HTML issue.
- Delivered within two business days after month end.
- Coverage gaps and soft-200 exclusions travel with the issue.
- Cancel before the next monthly compilation.
Sealed provenance
The source archive spans 2026-06-11 through 2026-08-07. The newest collection completed 2026-08-07T15:05:47.074463+00:00. Its universe binding is merchant_panel_v1.txt:dc47333a79a8b3f3 and its batch seal is 8c9c18a062836d1885e9de9b6307804328fa81f1012d83f81092bbd095656991. The paths observed are the public merchant URLs shown in the table above; each run is bound to the frozen panel and the row hashes recorded before any later adoption outcome is known.