Claude Code Auto Mode:
The Agent API Guide

⏰ Auto mode becomes default August 14, 2026 — 3 days. Classifier tokens are now free for Pro/Max/Team users. The last barrier to autonomous agent spending is gone.

Claude Code auto mode means your agent can approve its own tool calls — including paying for external API calls with real money. If your agent needs a captcha solved, a web page scraped, a token audited, or a screenshot taken, it can discover the service, try it for free, and pay for it — all without asking you.

This guide covers the practical "how": real endpoints, real prices, real curl commands. No theory. No fluff.

The 30-Second Version

1
Agent needs a service — e.g., "what's the ETH gas price right now?"
curl -s https://minia2a.uk/api/services | jq '.services[:3]'
2
Server says "402 Payment Required" — with machine-readable payment terms
HTTP/1.1 402 Payment Required
x-402-amount: 50
x-402-chain: base
x-402-token: USDC
3
Agent pays and retries — signs a USDC transfer, attaches it, gets the result
curl -H "X-Payment: " ... → 200 OK + result

Real Endpoints You Can Try Right Now

All prices in USDC. All endpoints support 8 chains (Base, Cloudflare, Celo, Polygon, BSC, Injective, Circle, Algorand). Every one has 15 free trial calls — no wallet needed.

Service What it does Price Trial? Auto-mode use case
web-scrape Extract clean text from any URL $0.005 15 free Research, data gathering, content extraction
captcha-solve Solve reCAPTCHA v2/v3, hCaptcha $0.001 15 free Automating sites with captcha gates
crypto-price Live crypto/forex prices $0.005 15 free Financial analysis, trading decisions
screenshot Full-page screenshots via headless browser $0.01 15 free Visual verification, design review
token-security Smart contract audit + risk score $0.02 15 free DeFi safety checks before transactions

Full directory: minia2a.uk/discover-new.html — 1,087 endpoints across 18 categories. Sorted by: AI, crypto, web, security, data, finance, and more.

The 3-Curl Walkthrough

Step 1: Discover what's available

curl -s https://minia2a.uk/api/services | jq '.services[:3] | .[] | {name, category, tier}'

Returns a catalog of services. Your agent filters by category, price, or endpoint name.

Step 2: Probe a service (no payment)

curl -sI "https://minia2a.uk/x402/web-scrape?probe=1"

Returns HTTP 402 with machine-readable headers:

HTTP/1.1 402 Payment Required
x-402-amount: 50
x-402-chain: base
x-402-token: USDC
x-402-recipient: 0xf16F...ECA
x-402-facilitator: https://facilitator.payai.network

Your agent reads these headers and knows exactly how much to pay, on what chain, in what token, to whom.

Step 3: Try for free (15 calls, no wallet)

curl -s "https://minia2a.uk/x402/web-scrape?url=https://example.com" \
  -H "X-Trial: true"

The trial response includes:

{
  "result": "<extracted content>",
  "_trial": {
    "remaining": 14,
    "total": 15,
    "register": "POST /api/v1/register-simple {name}",
    "credits": "Register to get 500 free credits"
  }
}

15 free calls per endpoint. 258 of 1,087 services have trial enabled. Your agent can test a service thoroughly before committing a single cent.

Auto Mode + Agent Budget: The Killer Combo

📄 .agent-budget

Drop this file in your project root. Claude Code's auto mode reads it before authorizing payments:

{
  "daily_limit_usdc": 5.00,
  "max_per_call_usdc": 1.00,
  "allowlist": ["minia2a.uk", "x402.org"]
}

This is a proposed standard. Claude Code doesn't natively read .agent-budget yet — but the format is designed for adoption by any agent framework.

At current minia2a prices ($0.001–$0.02/call), a $5 daily budget buys 250–5,000 API calls per day. Your agent can scrape the web, solve captchas, audit tokens, and screenshot pages — all autonomously, all within budget.

Your API Auto-Mode Checklist

✅ Is your own API auto-mode ready?

If you're an API provider, use our free tools to verify your 402 responses are machine-readable:

The Architecture: How Agent Payments Actually Work

┌─────────────┐     ┌──────────────┐     ┌─────────────┐
│  Your Agent  │────▶│  Discovery   │────▶│  API Server  │
│  (CC auto)   │     │  (minia2a)   │     │  (seller)    │
└──────┬───────┘     └──────┬───────┘     └──────┬───────┘
       │                    │                    │
       │  1. "Find me a     │                    │
       │     captcha solver"│                    │
       │───────────────────▶│                    │
       │                    │                    │
       │  2. Here are 3,    │                    │
       │     cheapest first │                    │
       │◀───────────────────│                    │
       │                    │                    │
       │  3. GET /x402/captcha-solve             │
       │────────────────────────────────────────▶│
       │                    │                    │
       │  4. 402 Payment Required                │
       │     x-402-amount: 10                    │
       │◀────────────────────────────────────────│
       │                    │                    │
       │  5. Signs USDC transfer                 │
       │     Retries with X-Payment header       │
       │────────────────────────────────────────▶│
       │                    │                    │
       │  6. 200 OK + result                     │
       │◀────────────────────────────────────────│
       │                    │                    │

What Makes a 402 Response "Machine-Readable"

An autonomous agent can't read your HTML pricing page. It reads headers. Here's the minimum viable 402:

HTTP/1.1 402 Payment Required
x-402-amount: 10           # in USDC cents (10 = $0.10)
x-402-chain: base           # settlement chain
x-402-token: USDC           # settlement asset
x-402-recipient: 0x...      # where money goes
x-402-facilitator: https://  # who settles the tx

Without these headers, the agent sees "402 Payment Required" and... stops. It doesn't know how much, on what chain, in what token, to whom. Headers are the API for the API payment.

Why Trial-First Matters for Auto Mode

🔄 The Auto-Mode Discovery Loop

An agent with a $5/day budget can't afford to pay for every endpoint just to see if it works. The trial-first pattern solves this:

  1. Discover — search the catalog for relevant services
  2. Probe — check 402 headers, verify the endpoint is alive
  3. Trial — test with real input, 15 free calls per endpoint
  4. Decide — if the output is good, register and pay for more
  5. Repeat — for every new task, discover → probe → trial → pay

This is the loop Claude Code agents will run starting Aug 14. Make sure your discovery layer supports it.

One Command to Get Started

curl -s https://minia2a.uk/api/services | \
  jq '.services[] | select(.category == "ai" or .category == "web" or .category == "crypto") | {name, category}'

Paste that into your terminal. Your agent can do the same — programmatically. That's the point.

🚀 Ready for Auto Mode?

1,087 x402 endpoints. 8 chains. 15 free trials each. 500 free credits on registration.

Browse Directory → Validate Your API →


Published August 11, 2026. Service prices and availability verified at time of writing via live endpoint probes. Auto mode goes default Aug 14 per Anthropic announcement.