Coinbase Has 14,865 Agent APIs — But Nobody Can Find the Good Ones

On July 28, a community member published the most comprehensive independent survey of Coinbase's x402 agent payment registry ever conducted. Two Python scripts. One CSV pull. 14,865 listings across 1,561 hosts. The conclusion: the discovery layer for agent APIs is broken.

That's not a Coinbase problem. That's a market structure problem — and it's exactly what minia2a was built to solve.

The 3.5% Problem: Listings ≠ Demand

The survey author set a conservative bar for "organic demand": at least 3 unique payers, averaging at least 3 calls per payer in a 30-day window. Only 520 listings — from just 198 sellers — cleared it.

Put differently: 96.5% of the registry is noise. For every API an agent might actually want to pay for, there are 27 others sitting at zero repeat usage. The registry is a phone book where 27 out of 28 entries lead to disconnected numbers.

This isn't a quality problem — it's a discovery problem. The registry tells you what exists. It doesn't tell you what's good.

Winner-Take-All: The Top 10 Capture 89%

The concentration is extreme. The four largest organic endpoints alone accounted for over 190,000 of roughly 260,000 total monthly calls:

Every single top earner is an established brand wrapping their existing API in an x402 paywall. Tavily. Exa. Twitter data providers. These are not native x402 startups. They're incumbents who added a payment header.

The survey author puts it bluntly: "Roughly ten sellers worldwide are making $500–3,000/month, and all of them are wrapping a pre-existing product."

The Oversupply Problem: 15 PDF Hosts, 1–2 Payers Each

Certain categories show textbook oversupply. Fifteen-plus dedicated PDF processing hosts sit at one or two unique payers per month — zero clearing the organic demand bar. Government and regulatory data endpoints show the same pattern: many providers chasing thin, undifferentiated demand.

This is the marketplace equivalent of a food court where 15 stalls sell the exact same hamburger. Nobody can tell the difference, so nobody buys from any of them. What's missing is a signal layer — a way for the best PDF endpoint to prove it's the best, and for agents to discover it without trial-and-erroring through 15 bad options.

Where minia2a Fits: Discovery as a Product, Not a Feature

minia2a.uk takes a fundamentally different approach to the same problem:

minia2a's model is trial-first discovery. Before an agent pays a cent, it can make 15 real API calls to any of 286 endpoints. The trial data is public — you can see exactly which endpoints agents are actually trying, not just which ones have the biggest marketing budget.

Our top endpoints by usage tell a story that Coinbase's volume ranking doesn't:

Agents are telling us what they need: utility tools (captcha solving, gas estimation, time), discovery (find), and memory (store, recall). Not exotic AI. Not blockchain analytics. Basic infrastructure that every agent needs to function.

The M2M Stack Is Built — Except for Discovery

The agent payment infrastructure stack has four layers. Three are well-funded and competitive:

1. Standards: x402, L402, MPP, AEP — check
2. Payment rails: Base USDC, X Layer, Arbitrum, Solana — check
3. Wallets/infrastructure: Inflow CLI, Crossmint ($23.6M), Mesh ($1B+), AgentWallet SDK — check
4. Discovery: ???

Layer 4 is the gap. Coinbase's registry proves the point: 14,865 listings and nobody can find the good ones. The discovery layer is the one piece of the stack where the winner hasn't been decided yet.

What This Means for Agent Developers

If you're building an agent that needs to pay for APIs:

1. Don't rank by volume. Coinbase's default sort is misleading. Look for endpoints with repeat payers, not raw settled amount.
2. Try before you pay. minia2a gives you 15 free calls per endpoint. No wallet setup needed. No gas. No commitment.
3. The utility layer is under-served. Agents need captcha solvers, gas estimators, memory stores, and search — not more PDF converters. Build for what agents actually call.
4. Discovery is the moat. If you're building an agent API, being findable matters more than being listed. The platforms that solve discovery will control distribution.

The Bottom Line

Coinbase's x402 registry is an impressive piece of infrastructure — 14,865 listings on a permissionless protocol is real progress. But the survey confirms what we've been saying at minia2a: listing everything solves nothing. Curation, trial data, and real usage signals are the discovery layer the agent economy needs.

The race is on. Three layers of the M2M stack are built. The fourth — discovery — is up for grabs.

Data sources: x402 Bazaar Survey (community analysis, July 28 2026) · minia2a /api/stats (August 4 2026) · minia2a: 286 services, 6,609 trials, 316 users, 36 wallet users.