For coding agents

Paste this into your agent's context, or point it at /llms.txt and the OpenAPI spec.

text

tinyx402: TinyFish web APIs paid per request via x402 (no API key).

Payment
  protocol   x402 v2, scheme "exact" (EIP-3009, gasless for payer)
  network    eip155:1874 (Whitechain Sepolia, RPC https://rpc.testnet.whitechain.io)
  asset      0xcedfB56113143d55b2358b8CFbB0dEB5Ad2f55a7 (USDT, 6 decimals, domain "Tether USD" v1)
  client     @x402/fetch wrapFetchWithPaymentFromConfig + @x402/evm/exact/client ExactEvmScheme
  safety     ALWAYS set spendControls.allowedAssets with maxAmountPerPayment

Paid endpoints (price in USDT base units)
  GET  /api/v1/search?query=&location=&domain_type=      10000  -> { results: [{ position, title, snippet, url, site_name }] }
  POST /api/v1/fetch     { urls[], format, purpose? }    15000  -> { results: [{ url, title, text }], errors: [] }
  POST /api/v1/agent/run { url, goal, output_schema? }   50000  -> { run_id, status, result }
  POST /api/v1/agent/run-sse   (same body, SSE)          50000
  POST /api/v1/research  { query, mode: deep|max }      200000 (max: 500000) -> { research_run_id, events_url }
  POST /api/v1/datasets  { prompt, target_rows }         20000 x target_rows -> { id, rows_url, events_url }
  POST /api/v1/browser   { url?, proxy_config? }         30000  -> { session_id, cdp_url }

Free follow-ups (IDs from the paid call)
  GET /api/v1/research/{id}            GET /api/v1/research/{id}/events (SSE)
  GET /api/v1/datasets/{id}            GET /api/v1/datasets/{id}/rows?after=   GET /api/v1/datasets/{id}/events?run= (SSE)
  DELETE /api/v1/browser/{id}

Rules
  - 4xx/5xx responses are never charged. Retry 502/503/504 once; do not retry 400.
  - Research and Datasets settle when the job is accepted; they then run 5-45 min in the background.
  - agent/run-sse sends the settlement as a final `payment` event, not a header.
  - Successful JSON responses carry a PAYMENT-RESPONSE header (base64 JSON with tx hash).

Picking an endpoint

You know the URL → fetch. Pass purpose, and use include_selectors to scope the extraction.
You need URLs → search, then fetch the best hits.
Interaction needed (clicks, forms, pagination) → agent/run, with an output_schema so the result is typed.
Open question across many sources → research, then follow events_url.
Many entities with the same fields → datasets, then read rows_url.
You want to drive the browser yourself → browser, then chromium.connectOverCDP(cdp_url).

Budgeting

A maxAmountPerPayment of 200000 (0.20 USDT) covers every fixed-price endpoint and deep research. Use 50000 to exclude research. Datasets cost 20000 × target_rows, so set the cap to fit. Agents can read the exact price from the 402 response before paying.