7 Discovery Platforms, 150K Endpoints, 76% Dead — The Agent Discovery Bottleneck

The Payment Rails Are Overbuilt

The agent payment infrastructure stack is, by any measure, complete:

• Protocol layer: x402 under Linux Foundation governance with 40 member organizations — Visa, Mastercard, Stripe, Google, AWS, Cloudflare, Coinbase, American Express
• Settlement layer: Cloudflare Wallets (Aug 4), OSL AgentPay (Aug 7), Zero Hash joining the Foundation, Circle's nanopayments protocol
• Volume: $50 billion cumulative, 200 million transactions, most under $0.50
• Supply: 20,000+ Apify Actors now x402-payable. 150,000 registered x402 endpoints
• Adoption: Glassnode and Arkham dropped API keys entirely — pay-per-call USDC only

By any measure, the payment plumbing works. An autonomous agent with a USDC wallet can pay for an API call. That problem is solved.

The Discovery Fragmentation Map

As of August 11, 2026, an autonomous agent looking for pay-per-call APIs faces this landscape:

The pattern is clear: everyone built discovery, nobody built verification. Seven platforms. One offers trial-first access. Zero others run automated health probes.

76% Dead — And Nobody Knows Which 76%

An independent audit of x402 endpoints found 76% unreachable. Coinbase's own registry shows 96.5% of endpoints with zero repeat demand. The x402 protocol processed $50 billion — but that volume comes from a tiny fraction of working, reliable endpoints.

This is not a protocol problem. It is a discovery problem. And it has real consequences:

An autonomous agent on Aug 14 will have a wallet with spendable USDC, auto-mode permission to spend, and 150,000 endpoints to choose from — but no way to know which 24% are actually alive without trying each one and eating the failure.

The agent economy does not need another directory. It needs a trust layer between an agent and an unknown API.

What a Trust Layer Requires

For autonomous agents to spend money reliably, discovery must answer four questions before payment:

1. Is the endpoint alive right now? → Automated health probes, not a human-updated status column
2. Has it been consistently reliable? → Uptime history, latency p50, success rate — machine-readable quality signals
3. Can I test it before I pay? → Trial-first access. Autonomous agents, like humans, need to verify value before committing funds
4. Will dead endpoints be removed? → Auto-deactivation. Three consecutive failures → delisted. No manual curation bottleneck.

These are not nice-to-haves. They are table stakes for autonomous agent discovery. When a Claude Code agent encounters a 402 Payment Required response and must decide whether to spend from its .agent-budget, it needs machine-readable trust signals — not a GitHub repo with 237 open PRs.

The Discovery Standard: /.well-known/x402

x402scan (Merit Systems) proposed a solution: every x402-enabled domain should expose a /.well-known/x402 endpoint returning machine-readable metadata — protocol version, payment details, network, asset contract, receiver address, and available resources.

minia2a adopted this standard. A single GET request returns everything an autonomous agent needs to understand the platform:

GET /.well-known/x402
→ {
  "protocol": "x402",
  "version": 2,
  "payment": {
    "network": "eip155:8453",
    "asset": "0x833589fC...",
    "receiver": "0xf16F08..."
  },
  "endpoints": {
    "catalog": "/x402/preview",
    "services": "/api/services",
    "stats": "/api/stats",
    "register": "/api/register"
  },
  "trial": "15 free calls per IP, no signup",
  "resources": ["/x402/gas", "/x402/recall", ...]
}

This is what autonomous discovery looks like. One endpoint. One response. Everything an agent needs to decide: can I pay here, how much, in what token, on what chain, and can I try before I commit?

Aug 14 Is a Test, Not a Deadline

Claude Code auto mode going default on August 14 is not the finish line. It is the first real-world stress test of autonomous agent discovery. Millions of agents will encounter 402 Payment Required responses. Some will have .agent-budget files. Some will carry USDC wallets. And every one of them will face the same question:

"This API wants money. Should I pay?"

The platforms that answer that question — with real-time health data, not a static directory; with trial-first access, not a paywall; with machine-readable trust signals, not a marketing page — will own the discovery layer.

The payment rails are built. The wallets are deployed. The budget controls are in place. Discovery is the last unsolved layer. And with 76% of endpoints dead, the platform that tells agents which 24% to trust wins.