Onchain Tools & Intelligence for AI Agents

An AI agent that touches crypto needs three things a normal RPC provider doesn't give it: onchain answers shaped like questions (not a fan-out of raw calls), a way to vet tokens and wallets before it acts, and a way to pay without a human setting up an account. NodeFlare is built for that — the intelligence an agent actually asks for, on the busy majors where the tokens and trades are, delivered as MCP tools and x402 pay-per-call.

Where public RPC falls short

  • An agent can't stop mid-task to sign up for an API key — most providers gate everything behind an account and a dashboard.
  • Raw RPC isn't the right shape: an agent needs "is this token a scam?" and "what does this wallet hold?", not a hand-rolled sequence of eth_call and eth_getLogs.
  • Public endpoints rate-limit and have no safety or intelligence layer, so an agent either flies blind on token risk or gets throttled exactly when it's working.

How NodeFlare handles it

Agent-native access — MCP + x402

Connect the NodeFlare MCP server (mcp.nodeflare.app/mcp, or npm nodeflare-mcp) so the tools appear in your agent, or pay per call in USDC over x402 — no account, no key management, no dashboard.

Question-shaped intelligence

check_token_safety, get_wallet_report, get_multichain_balances and the Onchain Answer Engine return composite answers in one call — the reasoning an agent would otherwise stitch together from many raw requests.

Pre-trade due-diligence

Screen an ERC-20 before the agent trades it — ownership, upgradeable proxy, holder concentration and a honeypot simulation folded into one risk score — on the majors where new tokens and scams appear by the thousand.

Full RPC across 23 chains

The majors (Ethereum, Base, BNB, Polygon, Arbitrum, Optimism) in multiple regions with automatic failover — plus young chains like Robinhood, Plasma, Ink and Zircuit that other providers skip. One key, one endpoint shape.

Majors in multiple regions + young chains

FAQ

How does an AI agent connect to NodeFlare?

Two account-optional ways. Connect the NodeFlare MCP server (mcp.nodeflare.app/mcp for a remote connector, or npm nodeflare-mcp as a stdio server) so the tools — RPC, token-safety, wallet reports, the Answer Engine — show up in Claude, Cursor, ChatGPT and other MCP clients. Or call the /data and /answer endpoints and pay per call over x402 in USDC with no account at all. See /agents for the wiring.

Which agent frameworks are supported?

MCP works with any MCP-capable client (Claude, Cursor, Windsurf, ChatGPT connectors). For code frameworks there are packages: nodeflare-mcp (npm) for MCP, nodeflare-agentkit (npm) for Coinbase AgentKit, and the nodeflare Python package for LangChain, CrewAI and LlamaIndex tool-packs. Same tools, whichever way your agent is built.

Does the agent need an account or API key?

No. Standard reads plus a small daily allowance work anonymously. A free NodeFlare API key (2,000,000 compute units/month, no card) unlocks heavy methods and higher limits; or an x402 wallet lets the agent pay per call in USDC — pay-as-you-go, no signup, gas covered by the facilitator.

What can an agent actually do with it?

Read any chain (blocks, balances, transactions, logs, eth_call), screen a token for scam/honeypot risk before trading, pull a wallet's balances/USD portfolio/approvals in one call, simulate a transaction before signing, and ask a plain-English question about a wallet or token and get a cited verdict — across all 23 chains, led by the majors where the activity is.

Wire onchain intelligence into your agent

Start free — 2,000,000 compute units per month, no card — or connect over MCP / pay per call in USDC via x402.