Which AI agent standards are actually adopted on the web?

All stats

In the week of August 17, 2026, of the 107,985 top web domains a Cloudflare Radar agent-readiness scan successfully crawled, only 3 serve x402 payments and exactly 2 advertises an AP2 agent-payment endpoint — the two most-hyped agent-commerce standards are, in raw web adoption, effectively at zero. The broadest signal, UCP, matches 8,788 domains (a suspected over-detection), while 324 publish an MCP server card and 38 an A2A agent card.

Updated as of (UTC) · source: agenteconomy.to/data.json

Agent-standard web adoption — key figures
Top domains scanned107,985of 160,188 crawled · data week 2026-08-17
UCP8,7888.14% of domains · MoM +3,240
MCP server card3240.300% of domains · MoM +45
web-bot-auth570.053% of domains · MoM +12
A2A agent card380.035% of domains · MoM +13
ACP110.010% of domains · MoM +4
MPP60.0056% of domains · MoM -1
x40230.0028% of domains · MoM +1
AP220.0019% of domains · MoM +2
WebMCP00% of domains · MoM ±0
Domains serving each agent standard (raw counts — shares not comparable) · domains
02.3K4.5K6.8K9KUCPMCP server cardweb-bot-authA2A agent cardACPMPPx402AP2
Domains serving each agent standard (raw counts — shares not comparable), in domains
Perioddomains
UCP8,788
MCP server card324
web-bot-auth57
A2A agent card38
ACP11
MPP6
x4023
AP22

Adoption here is a weekly scan, not a live counter

These figures come from a weekly agent-readiness scan of the top web domains, so the as-of stamp on this page is a data week, not a single day. A value that ticks up between Monday and Thursday is the same weekly measurement re-read; treating it as a daily series over-reads noise that is not there. Month-over-month change compares this week's scan against the scan from roughly a month earlier — the honest grain for a signal that moves slowly.

Why these shares cannot be compared to each other

Each check has a different meaningful denominator, so ranking standards by their bare share across all domains is a category error. x402 is a payment header for paid content and APIs — it only belongs on the small slice of sites that actually sell access — so its share against every crawled domain understates its penetration of the sites where it would ever appear. AP2, MPP and ACP are similarly niche. A crawl-control file that belongs on every website and a payment endpoint meant for a few thousand commerce sites are not on the same axis. The counts below are raw domain matches; the shares are shown for scale, never for cross-standard ranking.

The UCP number to distrust

UCP's count sits an order of magnitude above every other agent standard, and that gap is most likely a detection artifact rather than real adoption. The scan's UCP check reads as a loose heuristic that over-matches domains which do not actually implement the Universal Commerce Protocol. Treat the UCP row as a suspected over-count, and weight the smaller, stricter checks — MCP server cards, A2A agent cards, x402 — more heavily when reading genuine adoption.

Also asked

Does a low x402 domain count mean x402 is failing?

No — x402 is a payment standard for paid content and APIs, so it only belongs on the small subset of sites that sell access. Its share across all crawled domains understates its penetration of the sites where it would ever appear, and on-chain x402 settlement counts (millions of transactions) tell a very different story from web-domain presence.

Why is the UCP number so much higher than the others?

It is almost certainly a detection artifact. The scan's UCP check reads as a loose heuristic that over-matches domains, so the UCP row should be treated as a suspected over-count rather than real adoption of the Universal Commerce Protocol.

Cite this page

agent economy, “Which AI agent standards are actually adopted on the web?”, agenteconomy.to/stats/which-agent-standards-are-actually-adopted, figures as of August 18, 2026 (UTC). Recomputable from agenteconomy.to/data.json.

Sources
Keep digging