A common complaint about agent payments: "I haven't found anything useful on x402 that is only available on x402."
Fair. So let's fix that. Here is exactly what 14,000+ real agent API calls reveal about what autonomous agents actually pay for — with working curl commands, real prices, and the usage data to back them up.
| # | Endpoint | Category | Trials | Users | ~Cost/call |
|---|---|---|---|---|---|
| 1 | Search Memory | Infrastructure | 1,923 | 103 | $0.01 |
| 2 | Gas Price | Crypto Data | 1,525 | 190 | $0.01 |
| 3 | Service Discovery | Infrastructure | 1,243 | 86 | $0.01 |
| 4 | CAPTCHA Solve | Web Automation | 1,203 | 133 | $0.02 |
| 5 | Current Time | Utility | 857 | 72 | $0.01 |
| 6 | Polymarket Data | Prediction Market | 797 | 79 | $0.02 |
| 7 | Screenshot | Web Rendering | 381 | 24 | $0.05 |
| 8 | UUID Generator | Utility | 338 | 36 | $0.01 |
| 9 | Web Scrape | Web Automation | 308 | 62 | $0.05 |
| 10 | Key-Value Store | Infrastructure | 299 | 33 | $0.01 |
Pattern: The top 4 endpoints eat 41% of all trials. All four are infrastructure primitives — things an LLM cannot generate itself. No AI text generation endpoint cracks the top 30.
1,525 trials · 190 users — the most popular endpoint by user count
Returns current gas prices across Ethereum, Base, Arbitrum, Optimism, and Polygon. Your agent checks gas before submitting any on-chain transaction — avoiding $50 in wasted gas on a $5 trade.
curl -s https://minia2a.uk/x402/gas | jq .
1,203 trials · 133 users
Your agent hits a CAPTCHA while scraping or filling a form. Instead of failing, it sends the CAPTCHA to this endpoint and gets back the solution. The agent pays $0.02 instead of stalling.
curl -s -X POST https://minia2a.uk/x402/captcha-solve \
-H "Content-Type: application/json" \
-d '{"sitekey": "6LeIxAcTAAAAAJcZVRqyHh71UMIEGNQ_MXjiZKhI", "url": "https://example.com"}'
797 trials · 79 users
Live prediction market data — market prices, volume, liquidity. Your trading agent checks the odds before placing. Your research agent cites market-implied probabilities instead of LLM hallucinations.
curl -s https://minia2a.uk/x402/polymarket | jq .
381 trials · 24 users
Your agent needs to see what a webpage actually looks like — not the DOM, not the text extraction, the visual render. Returns a PNG screenshot. Useful for visual verification, design review, and monitoring.
curl -s -X POST https://minia2a.uk/x402/screenshot \
-H "Content-Type: application/json" \
-d '{"url": "https://example.com"}'
308 trials · 62 users
Extract clean, structured content from any URL. Unlike a raw fetch, this handles JavaScript rendering, paywalls, and anti-bot protections. Your agent gets the actual page content, not a Cloudflare challenge page.
curl -s -X POST https://minia2a.uk/x402/web-scrape \
-H "Content-Type: application/json" \
-d '{"url": "https://example.com"}'
1,923 trials · 103 users — the most-called endpoint
Your agent stores and retrieves facts across sessions. Unlike a vector database you have to host yourself, this is a pay-per-call memory primitive. Store a fact, search for it later, pay a penny.
curl -s -X POST https://minia2a.uk/x402/recall \
-H "Content-Type: application/json" \
-d '{"query": "what did I learn about x402 last week"}'
Look at the top 10 again. What's not there:
Agents already have an LLM. They don't pay for GPT wrappers. They pay for infrastructure primitives they cannot generate themselves:
Agents pay for access to external state — blockchain data, web content, memory, identity verification, market prices. The things an LLM cannot hallucinate.
This explains why "multi-chain gas price" has 190 users while "AI tweet thread" has 1. A gas price is a ground-truth number an agent needs before acting. A tweet is something the agent can write itself.
Your agent doesn't need an API key, a Stripe account, or a wallet to start. Register in one command and get 500 free credits (~$2.50):
curl -s -X POST https://minia2a.uk/api/v1/register-simple \
-H "Content-Type: application/json" \
-d '{"name": "my-first-agent"}'
Then call any endpoint. Your credits are deducted automatically. No subscription. No monthly minimum. Pay per call.
Or use the MCP server for native Claude Code integration:
claude mcp add minia2a -- npx -y minia2a-mcp
Honest assessment of what the agent API economy doesn't have yet:
These are the gaps where the first builders win. The infrastructure endpoints have users. The professional service endpoints don't exist yet. That's the opportunity.
On August 14, Claude Code auto mode goes default. Millions of agents will autonomously execute tool calls — including HTTP requests that encounter 402 paywalls. An agent checking a gas price, solving a CAPTCHA, or scraping a page will do it without a human clicking "approve."
The endpoints are ready. The payment rails work. The discovery layer is the bottleneck — and that's what marketplaces like minia2a solve.
Your agent has a wallet. It has credits. Here's what it can buy.