Who Buys x402 Tools
HTTP/1.1 402 Payment RequiredSeptember 2026~9 min read

I built a factory for x402 tools. Then I checked who's buying.

Five live tools, a build pipeline with more quality gates than customers, and zero external buyers. So I mapped every x402 payment on Base over 30 days to find out whether the market I was building for exists.

Jannis · data from x402scan's public API, snapshot 23 Sep 2026

90%of the money comes from buyers who pay exactly one seller
$10.7kspent in 30 days by buyers who shop across sellers, network-wide
0external buyers for my five tools

What I built

x402 turns the old 402 Payment Required status code into a real payment flow: an agent calls an endpoint, gets a price back, signs a USDC payment and retries. No account, no API key. The pitch is that AI agents will find tools on their own and pay per call.

I took that pitch literally and built a factory for such tools. An idea scout, a viability review with live web search, a two-pass build, an automated smoke test, QA, and a publish gate. The tools run on Cloudflare Workers, settle through a chain of three facilitators (Coinbase CDP, then PayAI, then Primer), sit in both Bazaar discovery catalogs, and ship with an MCP server on npm so agents can call them directly.

Five tools went live on Base mainnet at 1 to 3 cents a call: a markdown extractor, a SPA-to-markdown renderer, a burn-once secret vault, an SVG bar-chart renderer and a GeoTIFF transcoder. They've taken 68 paid calls so far, and all 68 came from my own wallets.

The question I should have asked first

"Build a good tool and list it in a catalog" only works if buyers shop around, meaning agents that discover sellers and pay whoever fits the job. The alternative is that each seller brings its own users, and x402 is just the payment rail inside a product that already has them. In that world, a catalog listing is a storefront on a street nobody walks down.

That question has a public, on-chain answer, so I went and got it.

How I measured it

x402scan indexes x402 settlements and its API is public (the app is open source). I pulled every buyer on Base in the last 30 days, 18,291 wallets, and for each one the sellers it paid. That gives the complete buyer-to-seller graph, not a sample. Where a seller uses several payment addresses on one website, I merged them into one operator.

Two filters came first, because neither is anyone buying a tool:

As a sanity check, my own payment address came out exactly right: three wallets, 68 transactions, all mine.

Finding 1: the market is captive

I sorted every buyer into one of three groups: captive (pays exactly one operator), open market (pays two or more, with repeat use) and probe (pays many sellers once or twice each, i.e. indexers and test wallets, like mine).

x402 buyers on Base by type, last 30 days
After filtering circular payments and transfers over $5 per call. 17,681 buyers, 2.18M transactions, $147.5k.
Show the numbers
TypeBuyersTransactionsUSDC
Captive15,772 (89.2%)1,904,905 (87.3%)$133,317 (90.4%)
Open market1,681 (9.5%)203,763 (9.3%)$10,741 (7.3%)
Probe228 (1.3%)73,847 (3.4%)$3,475 (2.4%)

Nine out of ten dollars come from wallets that pay a single operator all month. The biggest sellers by buyer count are almost entirely captive: ax1.vc has 2,620 buyers and 99.8% of them pay only ax1.vc; clusterprotocol took $58k, all of it from its own captive buyers. That pattern looks like apps that give their users wallets and route payments through x402, not agents browsing a catalog.

x402 today is mostly a payment rail inside products that already have users, not a marketplace where agents shop.

Finding 2: the open market is about $10k a month, for everyone

The buyers who do shop across sellers exist, but there aren't many and they don't spend much: 1,681 wallets, $10.7k in 30 days, across the whole network. The median one uses two operators and spends $0.47 a month.

Spread across 16,656 operators, that pool is thin. 89% of operators have zero repeat buyers from the open market. 145 earned $10 or more from it. My tools are in the 89%, together with almost everyone else.

Finding 3: who wins the open market

The operators that open-market buyers keep coming back to fall into three groups:

None of these are "a small clever tool, listed in a catalog." They either sell access with a real cost behind it, or they sit where agents' wallets already are.

Finding 4: the niche I liked is crowded and empty

My favorite tool is the burn-once secret vault, a neutral way for agents to hand over a secret exactly once, with a signed receipt. I thought "trust primitives for agents" (secrets, receipts, attestation, timestamping, escrow) was an open niche.

It isn't. At least two dozen sellers are already in it, and they're tiny: the biggest one I found has 48 buyers and $32 over 30 days, and most have fewer than ten buyers and under a dollar. Plenty of people had the same idea; the buyers never showed up.

Finding 5: and it's shrinking

Monthly x402 settlement volume on Base, USDC
Jul 2025 – Sep 2026 · $M
2026 only · $k, own scale
Gross volume as x402scan reports it, before any wash filtering. September covers 23 days. The 2026 panel has its own scale so the recent trend is visible.
Show the numbers
MonthUSDCAvg daily buyers

The spike in late 2025 was mostly the PING token mint: you paid via x402 and got tokens back, and tens of thousands of wallets did. Since then volume has drifted down, from about $2M a month in the spring to roughly $0.65–1M a month over the summer, and average daily buyers fell from about 6,000 in May to about 2,300 now.

Other analysts see the same direction with different absolute numbers, because they filter differently. An Artemis analysis found roughly half of x402 transactions were "gamed" (self-dealing or wash trading). Jamie Coutts, citing Helios data, reported settlement volume down 93% year to date. American Banker ran the story as "agentic payment rail shows volume decline."

What's still true

None of this means x402 is dead. The infrastructure keeps getting serious backing. The x402 Foundation launched under the Linux Foundation in April 2026, with members like Google, Visa, Mastercard, Stripe, AWS and Shopify. Cloudflare announced a Monetization Gateway in July (early access) that lets its customers charge for pages, APIs and MCP tools via x402 without writing payment code. Stripe runs its own Machine Payments Protocol and supports x402 too.

The forecasts are enormous: Gartner says AI agents could intermediate $15 trillion in purchases by 2028, McKinsey puts agentic retail commerce at $3 to 5 trillion by 2030. But those numbers are about agents buying things for people, not about agents browsing a catalog of small paid tools. Keyrock counted $73M across about 176M agent transactions on blockchain rails from May 2025 to April 2026. That's real, and it's tiny next to the forecasts.

The infrastructure has arrived before the economy it was built for.

What I'd tell someone starting an x402 project today

What's next for me

The five tools stay live, since they cost almost nothing. I'm not building tool number six into an empty market. The next experiment is small: make the tools visible in the places where agent wallets actually look, and re-run this analysis in 30 days to see whether anything moves.

If you're building on x402 and see different numbers, or you've found where the buyers are, I'd like to hear about it.

Method and sources