Economic-intelligence layer for AI agents: budget-capped tool calls with verifiable receipts, plus verified_route buyer-side x402 trust routing. 17 tools free to start, no keys; set AGENTPAY_BASE_KEY to settle paid tools in-place (gasless EIP-3009 on Base) under an AGENTPAY_MAX_SPEND cap.
AgentPay MCP exhibits significant definition quality gaps. Tool descriptions are present but often vague and marketing-focused rather than LLM-optimized. Input schemas are visible but lack proper type definitions and constraints. Critical issues: (1) descriptions like 'Usage-vetted pick of the best x402 provider' and 'Hard multi-call spend cap with a verifiable receipt ledger' are cryptic and assume domain knowledge; (2) many parameters lack type constraints (e.g., 'min_amount' in whale_activity is a string with no format/range guidance); (3) no output schemas documented; (4) duplicate tool names (gas_tracker appears twice); (5) no error handling guidance; (6) parameters like 'need' in verified_route are dangerously open-ended. The server reads as a domain-specific payment/trading tool, but definitions prioritize marketing language over LLM clarity.
Legacy alias for token_market_data - Get DEX liquidity data
Execute a Dune Analytics query and return live onchain results
Get funding rates for perpetual futures markets
Get current Ethereum gas prices and network status
Track gas prices and network congestion
Get open interest data for derivatives markets
Get orderbook depth and liquidity distribution
Cryptic, domain-jargon descriptions that assume deep x402/crypto knowledge. 'Usage-vetted pick of the best x402 provider' and 'Hard multi-call spend cap with a verifiable receipt ledger' are marketing copy, not LLM-actionable guidance. LLMs cannot infer when to call these tools or what they return.
Duplicate tool name 'gas_tracker' (appears at indices 3 and 11). This breaks tool discovery and forces LLMs to guess which variant to call. One must be renamed or removed.
Parameters lack type constraints and format guidance. 'min_amount' in whale_activity is typed as string with no format/range (is it wei? USDC? percentage?). 'need' in verified_route is a free-form string with no enum or pattern. This invites hallucinated values.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 51 | 2026-07-28+ | v2 |
Pre-trade risk check for AI agents trading crypto: live orderbook slippage at YOUR size, side-aware funding carry, open-interest crowding, and optional contract security — composed into a single ok/caution/avoid verdict with per-factor reasons and raw data embedded
Hard multi-call spend cap with a verifiable receipt ledger
Get DEX liquidity and market data for token pairs
Get current token price data from CoinGecko
Usage-vetted pick of the best x402 provider for any need — sybil tails collapsed, ready-to-pay challenge included
Get wallet balance for a given address and token
Track large wallet transfers and exchange flows
Scan for yield opportunities across protocols
No output schemas documented. LLMs cannot plan downstream tool calls or extract required fields (e.g., does dune_query return rows as array? What fields? Does session_create return a session_id?). This forces agents to guess.
No error handling guidance. Tools return no recovery hints (e.g., 'Token not found. Try token_price() with a different symbol.' or 'Insufficient balance. Check wallet_balance() first.'). Agents cannot self-correct on failures.