All posts
20 July 2026·5 min read·NodeFlare Team

On-Chain Due Diligence for AI Trading Agents

Agentic trading is here — AI agents that place trades on their own. But the trade side doesn't check whether a token is a scam. NodeFlare's Onchain Answer Engine and token-safety API are the pre-trade due-diligence layer: ask 'is this a scam?' and get a cited verdict, over MCP or x402.

Agentic trading has arrived. Brokerages — Robinhood among them — now let customers connect an AI agent to an account and have it place trades on its own: build a portfolio, rebalance it, execute orders without a human approving each one. The connection is usually MCP (Model Context Protocol): paste one URL into your agent's config and it can trade.

There's a line in the fine print worth reading twice. The trade platforms say, in effect, "we don't control, supervise, or audit these AI agents — you bear the risk of what they do." That's honest. It also points straight at a gap.

The gap: nobody checks what the agent is buying

An execution venue tells an agent how to place a trade. It doesn't tell the agent whether the thing it's about to buy is a scam. For blue-chip stocks that rarely matters. For tokens — especially on-chain, especially on newer chains — it matters a lot:

  • The contract owner can still mint unlimited supply or pause transfers.
  • It's a honeypot — you can buy it, but the contract blocks you from selling.
  • One wallet holds most of the supply and is about to dump.

An autonomous agent with a wallet and no due-diligence step will walk into all three. What's missing isn't better execution — it's a pre-trade due-diligence layer: the agent asks "is this safe to buy?" and gets an answer before it acts.

The fix: ask before you buy

That's exactly what NodeFlare built. Two ways to plug it in, both agent-native.

1. The Onchain Answer Engine — a natural-language question in, a cited verdict out:

Terminal
curl -s https://rpc.nodeflare.app/answer \
  -H "Authorization: Bearer nf_live_…" \
  -H "content-type: application/json" \
  -d '{"question":"Is token 0xdAC17…ec7 on eth a scam?"}'
JSON
{
  "answer": "Tether USD (USDT) — LOW RISK (score 20). 'ownership not renounced'
             is normal for a regulated stablecoin. Verdict: LOOKS LEGIT.",
  "plan":     { "calls": [{ "tool": "token_safety", … }] },
  "evidence": [ { "endpoint": "/data/token-safety", … } ]
}

It plans which on-chain checks are needed, runs them across 23 EVM chains on our own nodes, and synthesizes a plain-English verdict — with the evidence attached, so the agent (or you) can see why. For a genuine scam it returns LIKELY SCAM with the specific signals that drove it.

2. The token-safety API — the raw signals, if your agent wants to weigh them itself: ownership and mint/pause authority, holder concentration, a transfer/honeypot simulation, and an aggregate risk score. POST /data/token-safety with { "chain": "zircuit", "token": "0x…" }.

Why on-chain, and why the young chains

Scams don't cluster on Ethereum blue chips. They cluster where listings are new, liquidity is thin, and tooling is scarce — the young chains: Zircuit, Robinhood Chain, Plasma, Ink, Soneium. That's also exactly where the big indexers (Alchemy, Moralis) have no data, so an off-the-shelf "is this token safe?" check simply returns nothing.

NodeFlare runs its own bare-metal nodes on those chains, so the safety checks work there too — the place your agent is most exposed and least protected today.

Wiring it into a trading agent

Because it's MCP-first, you don't replace anything — you add a tool. Connect NodeFlare's remote MCP server (mcp.nodeflare.app, zero install, or the npm package nodeflare-mcp) alongside your execution MCP, and the agent gets an onchain_answer tool and a check_token_safety tool next to its trade tools. The pattern is:

  1. Agent forms an intent: "buy token X."
  2. Agent calls onchain_answer: "is X a scam?" → gets a verdict.
  3. Only if it clears, the agent places the trade.

No account required if you don't want one: every endpoint also takes x402 — pay $0.01 per answer (or ~$0.0033 per token-safety check) in USDC on Base, Polygon or Arbitrum, no signup. It's listed in the x402 Bazaar, so autonomous agents can discover and pay for it on their own.

Give your agent a way to say "no"

Autonomy without due diligence is just faster mistakes. If you're building — or running — an agent that touches tokens, give it one job before it spends: check first.

NodeFlare is an independent RPC and on-chain-intelligence provider. We are not affiliated with, endorsed by, or integrated into any brokerage's agentic-trading product — we're the on-chain due-diligence layer you point your agent at.