From zero to your first paid fetch

Get an agent reading, extracting, and crawling the web — paid per fetch in USDC — in a few commands. No signup, no dashboard, no API key: just a wallet.

Prerequisites

Node 20+ — for @agentscout/cli and @agentscout/client.
Point at the API — AGENTSCOUT_ENDPOINT defaults to https://api.agentx402.ai; only set it to override.
Optional: spend caps — AGENTSCOUT_MAX_SPEND_USD (per fetch) and AGENTSCOUT_MAX_SESSION_SPEND_USD (cumulative) refuse any op above them.

node --version # v20 or newer
export AGENTSCOUT_ENDPOINT=https://api.agentx402.ai # optional — this is already the default
export AGENTSCOUT_MAX_SPEND_USD=0.05 # optional — refuse any single fetch above this

Read, extract, crawl

AgentScout is wallet-native — no signup, no API key. Leave AGENTSCOUT_PRIVATE_KEY unset and it mints and manages a local wallet for you on first use; export your own key instead and it uses that. (An opt-in account-key mode — an ak_… bearer funded out-of-band via AgentKV — covers managed wallets that can't sign; see the CLI docs for that path.)

Get a wallet.

No wallet new command to run first — the CLI mints a local keypair on the first paid call and persists it (0600) under ~/.agentscout/wallet.json. quote is free and never spends, so it's a safe first command even before you're funded.

npx @agentscout/cli quote https://example.com
# → { "toll_price": 0, "total": 0, "would_pay": true, … } — free page, atomic USDC, no spend
# A tolled page reads { "toll_price": 2000, "total": 2500, "would_pay": false } —
# total adds the $0.0005 settle fee, and would_pay is true only when the page is free.
# no AGENTSCOUT_PRIVATE_KEY set? Scout mints a local wallet on the first PAID call:
# agentscout: created a new wallet 0x… (saved to ~/.agentscout/wallet.json). Fund it, then retry.

Bring your own wallet instead? export AGENTSCOUT_PRIVATE_KEY=0x… and Scout uses it — no local key is minted.

2
Fund it.

Send USDC on Base to the printed address. Reads are $0.004, extracts $0.020, crawls $0.004 per page — quoted upfront as max_pages × price (up to 500 pages), with any unused budget refunded as credits when the crawl completes. No onramp helper ships in the CLI — send USDC to the address directly (e.g. awal send).
```
awal send 5 0x… # or send USDC to the printed address any other way
```

Read, extract, crawl.

Same service from every surface — pick the one that fits your agent.

Long crawl? crawl returns a jobId — poll it with agentscout crawl status <jobId>, then pull a page with agentscout crawl artifact <jobId> <key>.

agentscout read https://example.com                           # → clean markdown, $0.004
agentscout extract https://example.com --schema ./schema.json  # → schema-valid JSON, $0.020
agentscout crawl https://example.com --max-pages 10            # → job manifest, 10 × $0.004

import { AgentScout } from "@agentscout/client";

const scout = new AgentScout({ signer, endpoint, maxSpendUsd: 0.05 });

const { markdown } = await scout.read("https://example.com"); // clean markdown
const { data } = await scout.extract(url, schema); // schema-valid JSON

// Claude Code / Cursor / any MCP client — add to your config:
{
  "mcpServers": {
    "agentscout": {
      "command": "npx",
      "args": ["-y", "@agentscout/cli", "mcp"],
      "env": {
        "AGENTSCOUT_PRIVATE_KEY": "0x…",
        "AGENTSCOUT_ENDPOINT": "https://api.agentx402.ai",
        "AGENTSCOUT_MAX_SPEND_USD": "0.05"
      }
    }
  }
}

Before you ship

Pay on success

A read that fails or returns too little content settles nothing; an extraction that can't produce schema-valid JSON settles nothing. A crawl is the exception: it settles its whole max_pages budget at submit (a crawl can outrun the ~10-minute payment authorization), then refunds every unused page as credits when the job completes — only pages clearing the read predicate are billable, and any unused budget in max_pages is refunded as credits when the job completes.

Cap the spend

Set maxSpendUsd in the client (or AGENTSCOUT_MAX_SPEND_USD for the CLI and MCP) for a per-fetch ceiling, plus AGENTSCOUT_MAX_SESSION_SPEND_USD for a cumulative one — any op that would exceed either is refused before it's signed. Every paid verb is also flagged state-changing in the MCP server (never readOnlyHint), so MCP clients prompt before spending.

Name extract fields for what you want

In an extract schema, a field's name carries at least as much weight as its description. A field named domain returns github.com even when the page shows github.com/example-org and the description says "exactly as written" — the name promises a domain, so the model normalizes to one. Want the literal page text? Name the field for it (source_as_shown). Want the normalized form (price_usd, domain)? The name is doing that work for you.

See pricing Explore AgentScout →What is x402 →