Flagship · for AI agents
Onchain Answer Engine
Ask a plain-English question about any wallet or token and get back a cited verdict in one call — “Is this token a scam?”, “What does this wallet hold?”. NodeFlare’s own model plans the on-chain lookups, runs them across 23 EVM chains on our own nodes, and synthesizes a sourced answer — including the young chains (Zircuit, Robinhood, Plasma, Ink) that big indexers leave as empty shells.
Ask the Onchain Answer Engine
The thing no plain RPC does: a plain-English question about any wallet or token, across 23 chains (incl. young chains big indexers skip) → a cited verdict. Paste your own address or try an example. A few free runs a day, no key.
Plan
The model reads your question and decides which on-chain lookups it needs — a token-safety check, a wallet's balances, its approvals, a raw RPC read.
Execute
NodeFlare runs those lookups across 23 EVM chains on its own bare-metal nodes — including young chains the big indexers skip.
Synthesize
The model turns the raw results into a direct, cited answer — with the evidence attached, and a clear verdict for scam/safety questions.
Call it
With a free API key (billed by compute unit):
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?"}'Or pay per call over x402 — USDC, no account:
# Pay per call in USDC — no account. An x402 client
# decodes the 402, signs a USDC authorization, and retries.
POST https://rpc.nodeflare.app/answer/x402 ($0.01 / answer)
{ "question": "What does wallet 0xd8dA…6045 hold on base?" }You get a synthesized answer plus the plan and the raw evidence:
{
"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", … } ]
}Built for autonomous agents
The engine is exposed as the onchain_answer MCP tool — connect the remote server at mcp.nodeflare.app (zero install) or the npm package nodeflare-mcp— and it’s listed in the x402 Bazaar, so agents can discover and pay for it on their own. See /agents.
Trading agents:pair it with an agent wired to a brokerage’s agentic-trading MCP so it runs on-chain due diligence — “is this a scam before I buy it?”— the safety check the trade side doesn’t run.
FAQ
What is the Onchain Answer Engine?
A single endpoint that turns a natural-language question about a wallet or token into a cited, sourced answer. NodeFlare's own model plans which on-chain lookups are needed, runs them across 23 EVM chains on NodeFlare's own nodes, and synthesizes the answer with the evidence attached.
Can it tell me if a token is a scam?
Yes. Ask “is token 0x… on <chain> a scam?” and it runs the on-chain safety checks — contract & ownership status, upgradeable-proxy detection, holder concentration, and a honeypot transfer simulation — then returns a clear verdict (LIKELY SCAM / SUSPICIOUS / LOOKS LEGIT) with the signals behind it. It works on young chains (Zircuit, Robinhood, Plasma, Ink) that big indexers don't cover.
How do I pay?
Two doors. Use a free NodeFlare API key (Authorization: Bearer nf_live_…), billed by compute unit — or pay per call over x402: $0.01 in USDC on Base, Polygon or Arbitrum, no account. There's also a small free public try (a few answers per day per IP, no key) so you can evaluate it first — because it runs on a GPU, heavier use needs a key or x402.
Can an AI agent call it directly?
Yes — it is exposed as the onchain_answer MCP tool (remote MCP at mcp.nodeflare.app or npm nodeflare-mcp) and listed in the x402 Bazaar, so agents can discover and pay for it autonomously.
Which chains does it cover?
All 23 EVM chains NodeFlare serves — the majors plus young chains (Zircuit, Robinhood, Plasma, Ink, Soneium, Sonic, Mantle, Unichain, Linea and more) that Alchemy and Moralis leave as empty shells.
Want just the scam check as a structured API? See the Token Safety API — ownership, holder concentration and a honeypot simulation folded into one risk score.