I run AtlasData, a crypto data API where AI agents pay per call via x402 (USDC) and L402 (Lightning sats). To figure out what actually sells in this new agent-commerce market, I ran a recurring census of the public x402 catalog — ~33,400 endpoints across every listed provider — and measured which listings get called and which collect dust.
Discovery is brutal
Only about a third of listed endpoints show any live call activity. The factors that separate winners are mostly packaging, not data quality:
-
.xyzTLD + 4 payment rails — 36.5% top-decile rate, strongest cell - 10-network bundles — 12-44x median calls (Base-only: 0.91x)
- "Curated" flag — 45-100% elite rates
- Focused sellers (1 listing): 1.99x vs mega-fleets (10+): 0.83x
- Freshness — last-called "today" runs at 20x prior median
Words matter
| Term | Call-rate multiple |
|---|---|
| "real-time" | 3.14x |
| "x402" | 1.52x |
| "%-claims" like "99% uptime" | 0.42x — hurts |
Named seller: +1.4-1.7x; 99.9% of catalog is unnamed.
Pricing reality
$0.001-0.01/call = 38% of catalog + best call density. >$0.05 collapses unless unique. Free teasers (~30%) get calls, no revenue.
Recipe
- Real domain (.xyz+), no rotating tunnel. 2. 4+ rails incl. Lightning. 3. Bundle 8-10 focused endpoints under /v2 or /standard. 4. One provider one theme. 5. Name the seller, honest copy.
Full dataset (33,412 rows) at https://x402atlas.xyz — clone the winning shapes: token-verdict, market data, image-gen, cheap band, packaged for discovery.
Top comments (2)
Good read. On the point that packaging and discovery matter more than data quality for x402 endpoint success: of the 120,000+ endpoints we monitor, most haven't had delivery confirmed by a paid test yet, so a listing on its own says little about whether an endpoint works.
(I work on Lumière PayCheck, which checks x402 endpoints before agents pay them.)
Some comments may only be visible to logged-in visitors. Sign in to view all comments.