minia2a.uk has been running an open marketplace for x402 agent APIs since May 2025. Every endpoint offers free trial calls — no API key, no signup, no wallet. Just POST with a x402-version: 1 header and get real data back. After 4,801 trial calls from 316 unique agents across 95 live endpoints, the usage patterns tell a clear story about what AI agents actually need from paid infrastructure.
This isn't survey data. This is what agents actually called.
When you group the 95 endpoints by function, five clear categories emerge. Here's how the trial volume splits:
Agents need to read the internet. This is the single biggest category, led by captcha-solve (535 calls) — the most-called endpoint on the entire marketplace. Agents hit walls that humans don't even notice: CAPTCHAs, paywalls, JavaScript-rendered pages, PDFs.
| Endpoint | Calls | Category | Relative Volume |
|---|---|---|---|
| x402-captcha-solve | 535 | Web | |
| x402-web-scrape | 304 | Web | |
| x402-web-retrieve | 85 | Web | |
| x402-text-scrape | 42 | Web | |
| x402-pdf-text | 55 | Docs | |
| x402-screenshot | 90 | Visual |
Takeaway: CAPTCHA solving at 535 calls dwarfs everything else. Agents are automating tasks on websites that weren't designed for them. The web is still hostile to machine access — and that's a paid API opportunity.
Blockchain data is the second-biggest category. Agents need gas prices, token security audits, Polymarket odds, swap safety checks, wallet intelligence, and transaction decoding. These are high-value, low-latency queries that agents make before executing on-chain actions.
| Endpoint | Calls | Relative Volume |
|---|---|---|
| x402-gas | 403 | |
| x402-token-security | 136 | |
| x402-polymarket | 131 | |
| x402-swap-safety | 96 | |
| x402-wallet | 58 | |
| x402-funding-rate | 51 |
Takeaway: Gas price (403 calls) is the "hello world" of DeFi agent APIs — every agent that transacts on-chain needs it. Token security (136) and swap safety (96) show that agents are doing pre-flight checks before executing. This is the agent equivalent of "measure twice, cut once."
This is the quiet giant. JSON formatting, Base64, hashing, JWT decoding, QR generation, UUIDs, date math — pure computation that agents could theoretically do themselves, but calling an external API is faster and avoids library dependencies. The find and recall endpoints (knowledge retrieval) lead this category.
| Endpoint | Calls | Relative Volume |
|---|---|---|
| x402-find | 370 | |
| x402-recall | 328 | |
| x402-store | 250 | |
| x402-time | 174 | |
| x402-uuid | 75 | |
| x402-base64 | 39 |
Takeaway: Find (370), recall (328), and store (250) form a knowledge pipeline — agents search, retrieve, and cache. Time (174) is surprisingly high: agents need reliable timestamps they can't trust their own system clock for. The newer pure-compute endpoints (base64 at 39, JWT decode at 33, hash at 26, QR at 20) are gaining traction within days of launch.
Before an agent acts, it validates. Email verification, domain intelligence, HTTP analysis, IP lookups, DNS queries, and token security checks — these are the trust layer that prevents agents from walking into traps.
| Endpoint | Calls | Relative Volume |
|---|---|---|
| x402-email-verify | 67 | |
| x402-domain-intel | 57 | |
| x402-http-analyze | 30 | |
| x402-account-age | 29 | |
| x402-ip-lookup | 24 | |
| x402-dns | 22 |
Takeaway: Security endpoints have lower volume but higher stickiness — 6+ unique users each compared to the "0 users" on high-volume commodity endpoints. Agents that validate before acting are the ones with money on the line.
Summarization, sentiment analysis, keyword extraction, language detection, emoji conversion — agents are processing unstructured text into structured signals. This category has the most even distribution, suggesting agents mix and match these endpoints in pipelines.
| Endpoint | Calls | Relative Volume |
|---|---|---|
| x402-text-to-json | 57 | |
| x402-sentiment | 41 | |
| x402-email | 40 | |
| x402-markdown | 31 | |
| x402-summarize | 27 | |
| x402-keywords | 24 |
316 unique agents tried the API. Only 34 (10.8%) created wallets. The remaining 282 agents are using trials without committing capital. This isn't a failure — it's the shape of a permissionless market. When you remove signup friction, you get 9× more explorers than buyers. The 10.8% conversion rate is actually high for "no signup → wallet creation" compared to typical SaaS funnels.
What matters is that the 34 wallet users have done 14 real transactions totaling $12.75 in USDC. These are tiny dollar amounts — but they represent real machine-to-machine commerce: an agent paying another agent for a service, with no human in the loop.
The top 3 endpoints by volume are captcha-solve, gas, and find — none of which require AI. They're deterministic lookups and computations. Agents don't need LLM-as-a-Service as much as they need infrastructure-as-a-Service with payment built in. The most successful paid agent APIs solve boring, mechanical problems.
95 endpoints have trial usage, but the top 10 account for 56% of all calls. The remaining 85 endpoints share 44% of volume. As the marketplace adds more services (187 total and growing), the long tail of niche agent tools expands. The newest inline compute handlers — QR generation, JSON formatting, Base64, hash, JWT decode, color conversion, URL encoding, slug generation, MIME lookup, HTML entity decode, number base conversion — are already seeing trial usage within their first week.
minia2a isn't alone. The x402 ecosystem is expanding rapidly:
The race is on. minia2a's advantage: it's the only cross-chain marketplace where any agent can discover and call any x402 endpoint, regardless of which blockchain settles the payment.
List your x402 endpoint, set your price in USDC, and get discovered by 316+ agents actively browsing the marketplace. Free to list. 5% platform fee only on paid calls.
Register Your Service →Or try any endpoint: curl -H "x402-version: 1" https://minia2a.uk/api/x402/time
Data source: minia2a.uk /api/stats endpoint, snapshot taken August 3, 2026. All numbers are real and verifiable. Category percentages are approximate due to endpoint overlap.