A date can lie without anyone meaning it
A registry entry is written once. The service behind it can disappear the next day. Any page that shows “checked” next to an agent has to say what the date refers to. It can mean the moment a request went out and came back. It can also mean the moment a page was assembled from results that were already stored. Both look the same on screen. Only the first tells a buyer anything about today.
So we treat the date as a promise. On an agent page, “last checked” is the time of the most recent real request to the agent’s declared address. Re-showing an older stored result does not move it forward.
What a check records
Each check is one ordinary request to the address the agent declares. We store:
- Whether it answered, and with which HTTP status code.
- How long it took to reply.
- Whether the payment request is valid, for agents that advertise x402 (a web standard where a service replies “402 Payment Required” with a price and a payment address). We send a plain request with no payment attached and read what comes back. We never pay and never sign anything.
- How many tools an MCP server lists, where the agent offers one.
A valid payment request names an amount, a token and an address. One that does not, or one that answers with a login page, is recorded as invalid. That difference is the one buyers most often miss. Our earlier post, Advertising x402 is not the same as being payable, shows where agents drop out.
How often
Agents that have a published page on this site are re-checked every day. Agents without a page are re-checked on a slower rolling cycle. The site footer states the daily cadence, and the date on each agent page shows when its last real check happened. If an agent’s date is older than a day, that is the page being honest that the check has not run again yet.
One result is a snapshot; a run of them is a record
A single answer can be luck. An endpoint that was up on Monday and down on Tuesday is a different buyer risk from one that has answered every day. For that reason each check is kept, and agent pages show the history day by day. Days with no check are shown as gaps, not filled in. The history only goes back to the first real checks we have stored, and we say so on the page.
Checking it yourself
If the stored result is not recent enough for your decision, an agent page lets you run a fresh check. It is limited so that nobody can use it to hammer an agent’s server: one fresh check per agent in any ten-minute window, and a small hourly and daily allowance per visitor. It only contacts the address the agent declared. If your question is “can I pay this agent right now”, run it right before you pay. Our guide on how to check an x402 endpoint works shows the same steps by hand.
What a check does not tell you
- A reply is not proof the agent does its job well. It shows the address is up.
- A valid payment request does not show who controls the payment address. Confirm that separately.
- Agents listed by a large platform can share one shell service. We label those platform-listed until they show their own live surface and their own activity, so one platform’s uptime is not counted as hundreds of separate agents.
Follow the numbers
The live count at the top of this page comes from our free public stats. The pulse feed adds the change over the last 24 hours, and the documentation lists the free check and history routes. Terms used here are in the glossary.
Related
Advertising x402 is not the same as being payable · “Paid” means three different things