The question agents actually ask

Trading agents don't ask "what's the price of TOKEN." Price is cheap. They ask the harder version: should I swap this, right now, on this chain, for this wallet?

That question has four sub-questions hiding inside it. Each one has an answer somewhere. Stitching them together used to mean a Bloomberg terminal, a Chainalysis seat, a DEX aggregator account, and a wallet analytics dashboard. Roughly $20k/year per seat, plus license fees, plus a human babysitting the screens.

Or you can pay $0.025 in USDC and get the same decision support in a single request loop.

Four endpoints, one verdict

Here's the composition. 

POST /token-risk-score      $0.005
POST /arbitrage-spread      $0.008
POST /bridge-rates          $0.004
POST /wallet-pnl            $0.008
                            ------
                            $0.025

Each call returns structured JSON. Each one is independently useful. Stacked, they answer the swap-safety question end to end.

The agent code is small: 

const probe = async (token, chain, wallet) => {
  const [risk, spread, bridges, pnl] = await Promise.all([
    x402.post('/token-risk-score', { token, chain }),
    x402.post('/arbitrage-spread', { token, venues: ['uniswap','curve','1inch','balancer'] }),
    x402.post('/bridge-rates',     { token, from: chain, to: ['base','arbitrum','optimism','polygon'] }),
    x402.post('/wallet-pnl',       { wallet, window: '30d' }),
  ])
  return verdict(risk, spread, bridges, pnl)
}

function verdict(risk, spread, bridges, pnl) {
  if (risk.score > 0.7)              return { swap: false, reason: 'risk score' }
  if (spread.best.slippageBps > 200) return { swap: false, reason: 'slippage' }
  if (bridges.best.feeBps > 50)      return { swap: 'wait', reason: 'bridge fee' }
  if (pnl.realized30d < -1000)       return { swap: 'human review', reason: 'losing streak' }
  return { swap: true, venue: spread.best.name }
}

Four boolean checks. About 800ms wall-clock, parallel. $0.025 paid, settled on Base, no API keys, no contracts, no procurement call.

Why this beats the terminal

A Bloomberg terminal is one bundle priced at $20k/year. You pay whether you use 5% of it or 95%. A trading agent probably uses 1.4% of the surface, the slice that answers swap-safety, and subsidizes the other 98.6%.

x402 inverts that. You pay per call, per endpoint, per question. If today's check needs only risk plus spread, you pay $0.013. Next week you add holder-concentration as a fifth signal, and you wire in one more endpoint without renegotiating anything.

Run the math at scale. 10,000 probes a day at $0.025 is $250/day, $91k/year. Worse than Bloomberg? Only if you have one operator. Run 50 parallel agents and Bloomberg blows past $1M in seat licenses while your x402 spend doesn't budge, because per-call pricing scales with usage, not headcount.

And agents don't need 100% of a terminal's surface area. They need a narrow slice, repeatedly, with structured output. That's what stacked x402 endpoints are. The terminal model exists because humans pay for optionality. Agents don't.

What composes next

The pattern generalizes. Any agent question that decomposes into independent factual lookups is a composition problem. Underwriting a loan. Screening a counterparty. Sizing a position. Picking a fulfillment carrier. Pricing a parametric insurance contract. Same shape every time: small endpoints, structured output, parallel calls, cheap enough that an agent gladly pays for four.

Edge-market's whole cluster is built this way on purpose. token-risk-score doesn't know what arbitrage-spread does. Neither one cares about wallet-pnl. The agent does the joining at call time. That's the right division of labor for the agent economy, because the agent is the only piece that knows what question it's actually asking.

The Bloomberg-shaped bundle is a fossil. Per-call composition is what replaces it.