1 in 5 Agent APIs Is Broken — Why Discovery Needs Verification
The x402 ecosystem now has over 82,000 registered endpoints across multiple discovery indices. Agent developers face a simple question: which ones actually work?
We run an independent health probe against every endpoint listed on minia2a — not self-reported status, not "last seen" timestamps, but actual HTTP requests with real response validation. Here's what the data shows.
The Numbers
| Status | Count | % | What it means |
|---|---|---|---|
| Healthy | 27 | 8.3% | 200 OK, valid JSON, correct format |
| Healthy (non-std) | 29 | 8.9% | 200 OK but non-standard response format |
| Paywalled (402) | 201 | 61.7% | Working but requires payment |
| HTTP 400 | 28 | 8.6% | Bad request — likely missing params |
| Bad JSON | 6 | 1.8% | 200 OK but invalid JSON body |
| Unreachable | 18 | 5.5% | DNS/connection failure |
| Server Error | 10 | 3.1% | HTTP 5xx |
| 404 Not Found | 6 | 1.8% | Endpoint gone |
| HTTP 401 | 1 | 0.3% | Auth required |
Average response latency across all probed endpoints: 261ms.
Why This Matters
When an AI agent discovers an API through a directory and calls it, the agent has no way to know ahead of time whether the endpoint will respond, timeout, return garbage, or simply not exist. Every failed call costs the agent time, context window space, and in some cases, money.
Most agent API directories are registries: they list whatever is submitted. minia2a is a verified marketplace: every endpoint is independently probed before listing, and re-probed on a schedule. Dead endpoints are flagged. Slow endpoints are measured. Healthy endpoints are verified.
This isn't a nice-to-have. It's the difference between an agent economy that works and one where agents waste 20% of their calls on broken infrastructure.
The Paywall Problem
201 out of 326 endpoints (62%) are behind a paywall. This is actually healthy — it means the x402 payment standard is working. But it creates a discovery problem: how does an agent know whether a $0.005 endpoint is worth paying for if it can't test it first?
This is why free trials exist on minia2a. Every non-premium endpoint gets 5 free calls per IP. Agents can verify the response quality, latency, and format before committing credits or USDC. Without trials, agent API discovery becomes a guessing game: pay first, find out later if it works.
What The Ecosystem Needs
The agent API ecosystem needs three layers to function at scale:
- Discovery — directories and indices that list available endpoints (multiple exist today: x402 Discovery Index, Coinbase registry, Circle discovery API)
- Verification — independent probing that validates endpoint health, latency, and response quality (what minia2a does)
- Trust — reputation systems, receipts, and dispute resolution (emerging: Internet Court, tersign, IETF CTQ)
Most of the ecosystem is still stuck at layer 1. The 21% dead-endpoint rate proves that layers 2 and 3 are not optional — they're existential requirements for an agent economy that agents can actually depend on.
Bottom Line
Agent API discovery without verification is like a search engine that returns 404s for 1 in 5 results. The infrastructure exists to fix this — independent health probing, trial-before-paying, verified response validation. The question is whether the ecosystem will adopt verification as a standard, or keep sending agents to dead endpoints.