The x402 Discovery Problem: Why 169M Payments Missed 9 MCP Servers

Martin Havel did something honest. He runs 9 public MCP servers — business registers, due diligence, sanctions lists, FX rates, VAT data, European company registries, real-estate data. All useful. All serving real data. He instrumented them, watched the logs, and reported what he found:

"x402 processed 169M payments. On my nine MCP servers: zero. Only the scouts arrived."

This matters because Havel is exactly the kind of developer the x402 ecosystem needs: real services, real data, real utility. If he's seeing zero payments, something is broken. But it's not what you think.

The problem isn't the payment rails. x402 works — 169 million transactions prove that. The problem isn't the wallets. Coinbase, Circle, AgentWallex, and self-custodial options exist. The problem isn't even the services. Havel has 9 of them. GPU-Bridge has 30. OrbisAPI has 2,800+. minia2a alone aggregates 180.

The problem is discovery.

The scout economy: maps without travelers

Havel reported that the only x402 traffic hitting his servers came from catalog crawlers — services like 402.ad-mcp-probe that systematically map the ecosystem, indexing which endpoints are free vs. paid. The catalog operator even charges 0.10 USDC per listing.

So the x402 economy currently looks like this: catalog crawlers mapping services → charging listing fees to service providers → so service providers can be... indexed by more catalog crawlers. It's a map-making industry with no travelers. Infrastructure indexing infrastructure.

Havel frames this as an ambiguous signal:

"Rails before the train — infrastructure and catalogs always precede paying traffic (how every payment network starts). Or rails for a train that will never arrive — the classic tech-bubble pattern of ecosystems indexing themselves with no end customer."

He's right to be cautious. But he's missing a piece.

What's missing: the marketplace layer

The agent payments stack has matured into distinct layers:

Protocol layer: x402, MPP — how payment requests are standardized
Wallet layer: Coinbase CDP, AgentWallex, self-custodial — how agents hold and spend USDC
Service layer: MCP servers, APIs — what agents pay for
Discovery layer: ???

That last one is the gap. Agents need to know what services exist, what they cost, and whether they're worth paying for — before they make an HTTP call and get a 402. A 402 response only works if the agent already found the endpoint. But how did it find it?

This is what minia2a.uk is built for.

By the numbers: discovery drives trials, trials drive payments

The funnel is real: 316 agents discovered services through minia2a, tried them for free, and 34 converted to wallets. That's a 10.8% trial-to-wallet rate — in line with SaaS freemium benchmarks. The discovery → trial → payment loop works. The missing piece isn't technology — it's aggregation.

Why individual MCP servers struggle

Havel's experience isn't surprising when you look at the economics of individual service providers:

1. Discovery requires aggregation. No agent is going to stumble across a Czech business register MCP server. But if that server is listed in a marketplace alongside 179 other services — organized by category, with free trials — agents browsing for "business data" or "company verification" will find it.

2. Trust requires reputation. An agent encountering an unknown endpoint has no reason to pay it. A marketplace provides context: usage stats, trial counts, wallet user numbers. Social proof that the service is real and delivers.

3. Trials close the trust gap. Havel doesn't want to enable x402 without seeing demand. But the demand won't appear without a way to test. minia2a's free trial credits solve this chicken-and-egg problem — services get exposure, agents get to verify quality before paying.

What Havel's experiment will find

Havel built a "gauge" — a pre-registered, 30-day experiment with six metrics to measure whether real paying demand exists. He expects the most likely outcome is calls_reaching_gate = 0. He's probably right.

Not because there are no agents with wallets. Not because his services aren't valuable. But because a standalone MCP server with no discovery surface is invisible to the agents that would pay for it.

The experiment would look different if his services were listed on minia2a, categorized under "Business Intelligence" or "European Company Data," with free trial credits available. Agents browse by category. They try before they pay. They convert when the value is proven.

The path forward

Havel is a rigorous thinker — his pre-registered experiment design with falsifiable metrics is exactly the kind of discipline the ecosystem needs. But the experiment is testing the wrong variable. The question isn't "will agents stumble across my server and pay?" — it's "what happens when agents can discover my server in a marketplace designed for discovery?"

The x402 ecosystem is real. 169 million transactions. $15M+ in volume. Stripe, AWS, Google, Mastercard all building on it. But the infrastructure is incomplete without the marketplace layer. Payment rails move money. Marketplaces move attention. Agents need both.

minia2a.uk is building the attention layer. 180 services and growing. If you're running an MCP server and wondering where the paying agents are — they're looking for you. They just can't find you yet.