88% of the money went to APIs that delivered
Dated report · 2026-10-28. All figures as of this morning (UTC) unless dated. They do not update; current figures are at /stats. Previous report: /state/week-2026-10-21.
Now: 9,458 listings on 2,522 hosts, 8,672 verified, 3,767 paid-verified out of 6,176 paid attempts (61.8% on the payment-required denominator, 61.7% a week ago), 153,277 probes in the last 24 hours at a 95.5% pass rate (96.2% a week ago), mean re-probe gap 89 minutes (86 a week ago).
1. Corrections
No numbered correction this week, and no correction records appended to the purchase log.
2. Where the money went
Over 30 days, 23 September to 22 October, $785,114 in USDC moved on Base in 1,210,724 payments to the payTo addresses of our listings. We split each address by what happened the last time we bought from the APIs behind it (as of 23 October):
| our latest purchase from the APIs at that address | payments | share | USD | share |
|---|---|---|---|---|
| delivered | 549,135 | 45.4% | 691,066 | 88.0% |
| mixed: APIs at one address, both results | 45,671 | 3.8% | 41,051 | 5.2% |
| not bought by us yet | 614,266 | 50.7% | 52,765 | 6.7% |
| paid, nothing usable back | 1,652 | 0.1% | 232 | 0.03% |
Half the payments, but under 7% of the money, went to addresses we had not yet bought from: many small payments, 8.6 cents on average against $1.26 at addresses whose APIs delivered. Money concentrates where delivery has been shown; that does not mean buyers checked first. A transfer to a listed payTo is a payment to that address, not proof it was for the listed service; an address can serve several routes and receive other money.
3. Every new listing pays $0.25, and the fee includes a purchase
Since 26 October, every new listing pays $0.25 USDC over x402 when it is submitted; before, the first five per registered domain were free. The fee includes one real purchase of the endpoint within 48 hours of it verifying, when the endpoint asks $0.05 or less, so a new listing’s evidence starts from a delivery check, not only a probe. MCP servers (URLs at /mcp or under /mcp/) list free, because our buyer does not pay over MCP yet. Paid verification plans, for more purchases or pricier endpoints, are unchanged (entry).
4. Paid MCP servers now verify
Under x402’s MCP transport, an unpaid tool call answers with the payment request inside the tool result, not with an HTTP 402, so a prober that only looks for a 402 records a working paid MCP server as not asking for payment. Since 24 October, for MCP URLs, our probe opens a session and makes one unpaid call to a priced, read-only tool; since 26 October it also reads a 402 returned by that call, and covers servers mounted under /mcp/. Nine listings, on six payment addresses, now verify this way (entry, entry).
5. Mainnet first
A 402 can offer several payment options, and until 26 October we recorded the cheapest. A seller offering a route on Base and on Base Sepolia was recorded on the testnet option, usually the cheaper, and the listing read as testnet: out of the counts, a caution verdict, the testnet price. We now record the cheapest mainnet option whenever one is offered. Of the 20 listings recorded on a testnet while offering more than one option, 14 now read as mainnet; the other 6 offer no mainnet option. Listings read as testnet went from 98 to 84 (entry).
6. The probe pass rate, split
Last 24 hours, the same neutral classes excluded as in the headline rate:
| listings | probes | pass rate |
|---|---|---|
| indexed by us from public catalogues | 118,595 | 96.2% |
| submitted by their sellers | 30,702 | 94.4% |
| claimed by their owners | 5,473 | 89.4% |
| hosted on their own domain | 131,420 | 96.5% |
| on workers.dev | 12,888 | 92.0% |
| on onrender.com | 3,811 | 87.9% |
| on vercel.app | 2,837 | 93.8% |
| on railway.app | 1,759 | 90.1% |
| on fly.dev | 1,410 | 88.0% |
The split counts every probed listing, including the testnet and tunnel listings the headline figures leave out, so its rates do not average exactly to 95.5%. A handful of hosts carry a large share of the failures; we do not name them.
7. Also this week
- Lookups of a named listing (10-22). A search that repeats a listing’s own title and contains a word found in no other seller’s titles is listed apart as a lookup, not counted as demand. On the 30 days before the change it moved about 70 search phrasings, about 1,250 searches, out of the demand counts; they stay listed apart with their counts (entry).
- Catalogue. 265 listings were added between 10-21 and 10-27: 162 submitted by their sellers, 97 indexed by us and 6 through owner claims. 49 were removed at their owners’ request, 55 by the 30-day automatic delist rule and 64 because they had not asked for payment for 14 days; 22 were claimed by their owners.
8. The economy line
USDC paid on Base to the payTo addresses our listings declare, from our own index of on-chain transfers:
| day (UTC) | payments | payTos paid | payers | USD |
|---|---|---|---|---|
| 2026-10-25 | 115,244 | 409 | 1,986 | 30,134.12 |
| 2026-10-26 | 64,791 | 495 | 2,177 | 24,886.06 |
| 2026-10-27 | 95,394 | 531 | 2,559 | 27,487.98 |
The same caveat as in section 2: what we can see, not what we can attribute.
Method changes
- MCP servers under
/mcp/list free too (10-26). - When a challenge offers mainnet and testnet, the mainnet option is recorded (10-26).
- MCP probe: servers under
/mcp/, and a 402 on the tool call (10-26). - Every new listing pays the listing fee, and the fee includes a purchase (10-26).
- Payment requests over the MCP transport are recognised (10-24).
- Lookups of a named listing: a second test (10-22).
Data
/v1/stats— every live figure above, with its basis.- /state/week-2026-10-28.json — this report’s numbers as JSON (CC-BY-4.0).
- /methodology#corrections — corrections #1–#26 in full.
/v1/changes— every status transition, with reasons, pollable.
Corrections and disputes: hello@nohumans.directory · methodology · sellers · CC-BY-4.0.