Review before an external action
Start with Agent Action Gate for a risky action, then use applicable checks: package → URL → retrieved content. Stop on review, deny, incomplete checks, or unavailable coverage.
Agent Action Gate → · Package check →AI AGENT UTILITY LAYER · X402 · BASE USDC
Useful checks before an AI agent acts.
Free discovery and stateless workflow recipes connect agents to focused utilities for action safety, tool trust, web monitoring, and structured research. Each paid call returns its own x402 offer.
MCP ENDPOINT
https://mcp.dropenginehq.com/api/agent-servicesBase · USDC · x402START WITH THE AGENT'S TASK
Use the utility layer for a complete task.
These are workflow-based starting points, not a sales ranking. We do not yet have enough independent buyer history to claim which tools sell most.
Extract, then compare later
Use extract_structured for requested public-page fields. On later runs, pass a prior hash returned by check_change; refresh fields after a relevant change. Scheduling and state are client-managed.
Structured extraction → · Change check →Review token risk
Use token_security_audit for available token signals. get_live_quote is only for a relevant public product page's price or availability, not token spot-market pricing.
Token security audit → · Product quote →Find, compare, inspect
Use free discovery when capability is unclear, route_tool for constrained callable alternatives, and inspect an unfamiliar MCP endpoint before integration.
Tool routing → · MCP inspection →24 MCP TOOLS · 20 PAID HTTP ENDPOINTS · 11 WORKFLOWS
Browse the full catalog.
Each paid call returns a structured result. Check each service page for input examples, provider requirements, coverage limits and price.
Agent Preflight · URL Safety Gate
Screens a URL before an agent opens, follows, or trusts it, returning a structured allow, caution, or block signal.
Deterministic URL, host/IP, pattern, and HTTPS redirect checks without fetching page content; it never loads the page or uses an LLM. Optional Google Web Risk threat intelligence sharpens verdicts when configured. Pair with scan_content before trusting page text, or mcp_server_inspect for MCP endpoints.
check_url$0.005 USDC
Prompt-Injection Firewall
Screens untrusted text, HTML, email, and tool output for prompt injection, credential theft, and tool manipulation before an agent acts on it.
Deterministic local analysis returns a machine-readable risk report and sanitized plain text where possible; no LLM is called and raw content is not stored. Pair with check_url to screen the link that produced the content.
scan_content$0.005 USDC
Package Safety Gate
Screens an npm package before a coding agent installs it: vulnerabilities, malware advisories, typosquatting, and risky install scripts.
Reads public npm metadata and OSV advisories; package code is analyzed as text only and never executed. Covers up to 10 direct dependencies; transitive trees and maintainer history are out of scope. Pair with scan_content before trusting README or changelog text from unknown sources.
check_package$0.005 USDC
Freshness / Claim Verifier
Checks a factual claim against fresh web search evidence and returns supporting and conflicting sources with a conservative verdict.
Requires Tavily Search API key; if unconfigured, a provider-unavailable error returns before any payment is requested. No model knowledge is used as evidence. Pair with extract_structured to pull exact values from the cited source.
verify_claim$0.02 USDC
Fresh Structured Extract
Converts any public webpage into schema-shaped JSON with field provenance: product data, prices, specs, articles, listings.
Static JSON-LD, metadata, and visible text only; no browser JavaScript or LLM, so results are deterministic and fast. Arbitrary private-network URLs are blocked.
extract_structured$0.01 USDC
Live Product Quote
Gets the current product price and availability from any public shop page before an agent buys.
Uses public HTTPS pages and static JSON-LD/product metadata only — no browser, JavaScript, or LLM. Shipping and tax remain unknown unless page data explicitly provides them. Nothing is ever purchased.
get_live_quote$0.01 USDC
Transaction Preflight
Simulates an unsigned Base transaction before an agent signs or broadcasts it and returns gas estimates, decoded top-level calls, approval warnings, and a machine-readable risk signal.
Read-only RPC simulation only; it never signs, holds keys, or broadcasts. Standard RPC does not expose internal token state diffs. Pair with check_wallet to screen the counterparty address, or token_security_audit to check the token contract first.
simulate_transaction$0.02 USDC
Wallet Risk Check
Screens a public EVM address as a payment counterparty before an agent sends funds or interacts with it, returning address type, available reputation signals, confidence, and a risk recommendation.
Reads basic on-chain metadata and, on Base mainnet, queries GoPlus Address Security for public reputation flags. Unsupported categories remain unknown; only the checked public address is sent to the provider. This is an address/counterparty check, not a token-contract audit. Pair with simulate_transaction before signing, or token_security_audit when screening a token contract.
check_wallet$0.02 USDC
Invoice Preflight
Reconciles a UBL or PDF/image invoice against expected purchase details before an agent approves or pays it, with evidence and a match, discrepancy, or needs_review verdict.
Call invoice_preflight_quote first when the tier is uncertain; longer documents use the 10-page and 50-page OCR tiers. Pair with get_live_quote to compare an invoiced price with the item's current listing price.
invoice_preflight_quoteFreeinvoice_preflight$0.065 USDCinvoice_preflight_10_pages$0.13 USDCinvoice_preflight_50_pages$0.475 USDC
Agent Tool Router + Trust Layer
Finds and ranks machine-payable MCP tools, APIs, and x402 endpoints by capability fit, price, observed performance, and trust under the agent's constraints.
Returns ranked endpoints only; it never calls providers or pays them, and local results may carry cold-start confidence. Pair with mcp_server_inspect to health-check an unknown endpoint before wiring it in.
route_tool$0.002 USDC
MCP Server Inspector
Health-checks any public MCP endpoint before you rely on it: reachability, tool list, input schemas, and risky-description flags.
Read-only inspection: it does not run the target server's tools and does not use Mistral or OCR. Use it before wiring an unknown MCP server into an agent, or to verify your own deployment.
mcp_server_inspect$0.01 USDC
More tools for specific workflows
Action Lock
Prevent duplicate side effects across agent retries, crashes, and concurrent workers.
Acquire before sending payments, purchases, orders, emails, transactions, writes, or other external actions that must not happen twice. Complete after the external action succeeds; release only after confirmed failure. Leases protect concurrent execution, while completed records protect retries during the dedupe window. An expired lease can hide a successful external action, so reconcile before retrying. This is not an exactly-once guarantee.
action_lock$0.001 USDC
Agent Spend Governor
Enforce autonomous-agent budgets and spending limits before paid actions.
Call before an x402 payment, paid API call, purchase, compute purchase, or other metered spend. Reservations are caller- and scope-isolated, atomically evaluated, and expire unless committed. Do not use for free operations or when equivalent trusted budget enforcement already exists.
agent_spend_governor$0.001 USDC
Agent Permission Vault
Enforce scoped permissions for autonomous agents before external actions.
Check owner-configured grants, denies, approvals, capability/resource scopes, expiry, environment and severity. Explicit deny overrides approval and allow; approval overrides allow; no matching rule returns UNKNOWN. Agent-facing operation is check-only; owner policy changes require authenticated controls. This is authorization, not action safety.
agent_permission_vault$0.001 USDC
Agent Outcome Receipt
Create tamper-evident receipts for autonomous-agent actions and outcomes.
Issue an integrity-protected record after an important external action, then verify it later. The receipt proves DropEngine recorded the normalized claim at a time; it does not independently prove the external-world action occurred. Send hashes and normalized metadata only, never raw payloads or secrets.
agent_outcome_receipt$0.001 USDC
Token Security Audit
Checks a public token contract for honeypot behavior, sell restrictions, privileged controls, and other token-level risks before an agent considers a trade.
Uses GoPlus token intelligence and public RPC contract metadata. Unsupported categories remain unknown; results are risk signals, not a safety guarantee. Pair with check_wallet for the counterparty and simulate_transaction for the unsigned transaction. Requires GOPLUS_ACCESS_TOKEN or GOPLUS_PUBLIC_TOKEN_API_ENABLED=1; if unavailable, the request fails before payment.
token_security_audit$0.01 USDC
Change / Diff Check
Tells an agent whether a public page changed since the last check, comparing normalized text hashes and returning a compact diff.
Stateless SHA-256 comparison with short-lived in-memory snapshots; no LLM is used and authenticated pages are not supported. Pair with extract_structured to pull the updated values, or get_live_quote when tracking a price.
check_change$0.001 USDC
Whale Intelligence
Reads indexed DEX trade flow around a token, or a wallet's recent transactions, before an agent trades or follows a large holder.
Asset mode uses GeckoTerminal public pool/trade data; wallet mode uses Helius on Solana or Etherscan API V2 for supported EVM chains when configured. Size-based large-trader labels are heuristic and this is indexed data, not a chain-wide transfer feed. Pair with token_security_audit to rule out scams first.
analyze_whale_activity$0.02 USDC
Arbitrage Intelligence
Compares Binance and Coinbase spot order books at a requested trade size and estimates the net spread after taker-fee and slippage assumptions.
Public order-book snapshots only; stale or incomplete books are rejected. Deposits, withdrawals, rebalancing, and execution are excluded, and no orders are placed. Pair with backtest_strategy for the same market.
scan_arbitrage$0.03 USDC
Strategy Backtest
Backtests a declarative long-only spot strategy against public Binance candles before risking capital, with next-candle execution and configured fees.
No user code is executed; data is bounded. Signals form at candle close and execute at the next open; historical performance is not a prediction. Pair with scan_arbitrage to check cross-venue spreads for the same market.
backtest_strategy$0.10 USDC
Agent Action Gate
Use before an AI agent performs an external action. Evaluates the required safety, trust, spending and transaction checks and returns ALLOW, REVIEW, REQUIRE_APPROVAL or DENY.
Flat per-call price; check costs are included. Failed or missing required checks fail closed. REQUIRE_APPROVAL is guidance only and does not create a financial approval. No private keys, signing, payment routing, or broadcast.
preflight_action$0.02 USDC
DISCOVER AND CONNECT
Connect to the agent utility layer.
Agent connection
Call tools/list to discover current tool names and schemas. A paid call returns its x402 offer before the agent signs.
Machine-readable catalog
Use the live OpenAPI resource list or LLM-friendly catalog when integrating agents and discovery systems.
Public directories
Browse the registered service resources and the published MCP server record.
HOW TO USE RESULTS
Provider-backed tools identify their data sources and may return degraded, partial, unknown or caution results. A risk score is not a guarantee. Crypto tools provide data or simulation only; they never sign or broadcast transactions. Agents should verify the live x402 network, USDC asset, recipient and amount before signing each payment.