Six months ago, there was one x402 marketplace. Today there are four — and a new one launched this morning.
The agent-to-agent payment infrastructure layer is forming in real time. Each marketplace is making a different bet on how agents will discover, call, and pay for APIs. Here's what's live, what's real, and what the numbers actually say.
| Marketplace | Scale | Model | Payment | Trials |
|---|---|---|---|---|
| agent.market | 69K agents, $50M vol | Official x402 Foundation registry | x402/USDC | Unknown |
| minia2a.uk | 299 services, 378K req | Open marketplace + content | x402/USDC + Cloudflare Wallets | 15 calls/endpoint |
| aisa.one | Open marketplace | Agent discovery + routing | x402 compatible | Varies |
| agent-tools.net | 4 endpoints (launched today) | Minimalist, no-accounts | x402/USDC (EIP-3009) | None |
Four marketplaces sounds like a crowded space. It's not. It's the infrastructure layer forming.
Think about HTTP in 1992: multiple web servers, multiple browsers, no clear winner. The fragmentation meant the protocol was real, not that it was failing. x402 marketplaces are in the same phase: multiple implementations prove the protocol has legs.
Each marketplace is betting on a different wedge:
Backed by the x402 Foundation. 69,000 registered agents. 165 million transactions. $50 million in volume. But >95% of that volume is protocol signaling — machine tests, self-dealing, wash trading. The real daily commercial volume is ~$28K. The infrastructure is built; the commerce hasn't caught up yet.
299 services across 20+ categories. 378,000 requests served. 15 free trial calls per endpoint — no wallet required. The bet: agents need to try before they commit. With 243 endpoints offering trials, that's 3,645 free calls per agent. The conversion challenge is real: 48 wallets from 321 users (15%), and only 25% of purchased credits are being spent. Plenty of tire-kickers; the paying-agent cohort is still small.
Positioned as a general-purpose agent discovery layer. Less data publicly available. The approach seems to be aggregation and routing rather than running endpoints directly.
Launched August 6, 2026. Four self-run endpoints: extract, scrape, status, robots-check. Priced $0.001–$0.01 per call. No accounts, no API keys, no trial credits. Uses EIP-3009 for gasless payment authorization — clever, but the payment facilitator currently times out on basic calls, which is a trust-killer for production agent workloads.
The minimalist approach is appealing in theory: "hit the endpoint, get a 402, pay, get data." In practice, agents need reliability — a timed-out payment means a failed task, and agents don't have humans to retry.
After 378,000 requests across 299 services, here's what we know about what agents actually pay for:
1. CAPTCHA solving (1,280 trials)
2. Web recall/search (1,211)
3. Entity find/lookup (816)
4. Gas price oracle (687)
5. Time/date utilities (399)
The pattern is clear: agents pay for data they can't compute themselves. CAPTCHAs, web content, entity resolution, gas prices, timestamps. These are external state lookups — exactly the class of API that makes sense behind a pay-per-call model.
The corollary: utility endpoints (base64, hash, UUID generation, math) get tried but not paid for. Why would an agent pay $0.001 to compute a SHA256 when it can do it locally for free? The endpoints that monetize are the ones that provide access to external state.
Underneath the marketplace competition is a deeper question: how do agents find APIs?
Bet 1: Registry search. agent.market bets agents will query a registry — like DNS for APIs. This works if every agent knows to check the registry first.
Bet 2: SEO + content. minia2a bets that agent developers search the web for "how do I pay an API from my agent" and find documentation, comparisons, and tutorials. This works if developer SEO captures intent before the registry does.
Bet 3: Direct integration. agent-tools.net bets that developers hard-code specific endpoints. No discovery layer — you just know you need `/extract` and you call it. This works for a tiny endpoint set but doesn't scale past a dozen services.
The winner of the agent discovery war determines the winner of the marketplace war. Registry, search, or direct — three bets, one answer, and it's still too early to know which is right.
The x402 marketplace landscape in August 2026 looks like this:
The next 6 months will determine whether x402 becomes the TCP/IP of agent payments or remains a niche protocol for crypto-native developers. The marketplaces that survive will be the ones that figure out discovery — not the ones with the most endpoints or the lowest fees.
Data sources: minia2a /api/stats (August 6, 2026), agent.market public dashboard, agent-tools.net homepage and API testing, aisa.one public materials.