@agentgate-builder/tokenverdict

TokenVerdict: Base token safety check (honeypot, taxes, owner powers) for AI agents (x402 + MCP)

Is this Base token safe to buy? TokenVerdict checks one ERC-20 token on Base mainnet and returns SAFE / CAUTION / DANGER with a 0-100 score, a confidence level, and the reasons behind it:

• Sellability / honeypot: a simulated buy, wallet-to-wallet transfer and sell against the token's real pool (one eth_call with state overrides; nothing is ever sent on chain). Covers Uniswap v2 forks, Aerodrome classic, concentrated-liquidity pools (Uniswap v3, Aerodrome Slipstream, PancakeSwap v3, Algebra) and Uniswap v4 (native ETH or ERC-20 quote). Returns buy, transfer and sell tax and round-trip loss.
• Owner powers: mint, blacklist, pause, tax and limit setters, trading switches, balance edits, upgradeable proxy (EIP-1967 / beacon / legacy), renounced or contract owner. Uses the verified ABI, or bytecode selectors when the source isn't verified.
• Liquidity: depth across pools (DexScreener plus on-chain Uniswap v2/v3/v4, Aerodrome and PancakeSwap lookups) and LP burn on the main v2-style pool.
• Holders: top-10 concentration, with pool balances backed out (GeckoTerminal snapshot).
• Verified source (Sourcify) and age (exact contract creation block, found on-chain).

Built for trading agents, wallets, Telegram/Discord bots and anyone about to buy a token they found five minutes ago.

• Hosted API: https://api.youraitools.co/tokenverdict (also at https://tokenverdict.agentgate.workers.dev)
• Prices (USDC on Base over x402, no API key, no signup): /v1/lite $0.005 (verdict + reasons), /v1/quick $0.015 (full report with evidence), /v1/deep $0.05 (multi-pool simulation, pool table, creation block, GeckoTerminal cross-check). Invalid addresses, non-contracts, non-ERC-20s and source outages are never charged.
• Free: /v1/preview (verdict, score, flag counts; 10/min per IP), a shareable card at /t/<token> and a remote MCP server.
• Start here (all our tools, MCP configs, code samples): https://agentgate-source.pages.dev/start/

Honest limits

This is automated, informational risk screening, not financial advice, not an audit, no guarantee. A token that passes today can be changed tomorrow by whoever controls it (upgrades, tax changes, blacklists, liquidity pulls). Specifically:

• The simulation runs at the current block with a small trade size; time-based traps, per-wallet limits, cooldowns, block-number or tx.origin tricks and sells that only fail later or above some size can pass.
• Owner powers are detected from function names/selectors; obfuscated logic can hide them, and their presence doesn't mean they will be used.
• LP lock is only confirmed for burned LP tokens on v2-style pools; third-party lockers and concentrated-liquidity positions aren't checked.
• Market data comes from free public APIs that rate-limit. When they don't answer, TokenVerdict falls back to on-chain pool lookups (main DEXes, WETH/USDC pairs only) and marks the result degraded: true with lower confidence. Holder concentration may then be unknown.
• Base mainnet only.

MCP (Claude Desktop, Cursor, Windsurf, any MCP client)

Remote (free preview tier):

{ "mcpServers": { "tokenverdict": { "type": "http", "url": "https://api.youraitools.co/tokenverdict/mcp" } } }

Local stdio (this package):

{ "mcpServers": { "tokenverdict": { "command": "npx", "args": ["-y", "@agentgate-builder/tokenverdict"] } } }

Tool: check_token(token). Free by default (verdict, score, flag counts). For the full quick report on every call ($0.015), set TOKENVERDICT_X402_PRIVATE_KEY to a Base wallet key holding a little USDC, and add the optional peers:

{ "mcpServers": { "tokenverdict": { "command": "npx", "args": ["-y", "-p", "@agentgate-builder/tokenverdict", "-p", "@x402/fetch", "-p", "@x402/evm", "-p", "viem", "tokenverdict"], "env": { "TOKENVERDICT_X402_PRIVATE_KEY": "0x..." } } } }

The stdio server only pays the TokenVerdict pay-to address, on Base, at most $0.02 per call.

Docker: docker build -t tokenverdict-mcp . && docker run -i --rm tokenverdict-mcp

HTTP

# free preview
curl -s "https://api.youraitools.co/tokenverdict/v1/preview?token=0x532f27101965dd16442E59d40670FaF5eBB142E4"
# paid (any x402 client): GET ?token= or POST {"token": "0x..."}
curl -i "https://api.youraitools.co/tokenverdict/v1/quick?token=0x532f27101965dd16442E59d40670FaF5eBB142E4"   # 402 + PAYMENT-REQUIRED

import { wrapFetchWithPaymentFromConfig } from '@x402/fetch';
import { ExactEvmScheme } from '@x402/evm';
import { privateKeyToAccount } from 'viem/accounts';
const pay = wrapFetchWithPaymentFromConfig(fetch, { schemes: [{ network: 'eip155:8453', client: new ExactEvmScheme(privateKeyToAccount(process.env.BUYER_KEY)) }] });
const r = await (await pay('https://api.youraitools.co/tokenverdict/v1/quick?token=0x...')).json();
if (r.verdict !== 'SAFE') console.log(r.verdict, r.score, r.reasons.map((x) => x.text));

Library (free preview):

import { preview } from '@agentgate-builder/tokenverdict';
console.log(await preview('0x532f27101965dd16442e59d40670faf5ebb142e4'));

Response (quick, abridged)

{ "token": "0x826f…adeea", "symbol": "AYB", "verdict": "DANGER", "score": 0, "confidence": "high", "degraded": false,
  "reasons": [{ "severity": "danger", "code": "honeypot", "text": "HONEYPOT SIGNAL: a simulated buy worked but selling back failed (reverted: balance)." }, "…"],
  "checks": { "sellability": { "simulated": true, "buy_ok": true, "sell_ok": false, "buy_tax_pct": 0 }, "owner": {}, "powers": {}, "proxy": {}, "contract": {}, "liquidity": {}, "holders": {}, "age": {} },
  "limits": ["…"], "disclaimer": "Automated, informational risk signals … Not financial advice, not an audit, no guarantee …" }

Self-hosting

Cloudflare Workers (free plan works): npm install && npx wrangler deploy. Set PAY_TO, prices and FACILITATOR (payai default, or cdp with CDP_API_KEY_ID / CDP_API_KEY_SECRET secrets) in wrangler.toml. The simulator contract is in contracts/TVSim.sol (npm run build:sim regenerates src/simcode.js with solc 0.8.26). Tests: npm test (offline, replaying recorded Base mainnet reports).

MIT. Part of the AgentGate tools by YourAITools: https://youraitools.co/built-by-us