Agentrisk MCP server

AgentRisk M2M 🛡️

Pre-Trade Security Layer for Autonomous DeFi Agents on Base. Stop letting your autonomous trading bots get rekt by stealth honeypots, malicious taxes, and rugpulls.

⚡ The Problem

Autonomous trading bots are fast, but they are completely blind. Every minute, new unvetted tokens launch on Base. A significant share are deliberately engineered traps designed to block sells, drain executing wallets, or alter transfer taxes at the worst possible moment. Static blocklists (like standard free APIs) are useless against freshly deployed, obfuscated contracts. By the time human-curated lists update, your bot's capital is already gone.

🎯 The Solution

AgentRisk is an isolated, machine-readable security API built specifically for AI agents and automated scripts. Before your bot executes a swap() on Uniswap or Aerodrome, it pings AgentRisk. We run a deep runtime simulation combining direct on-chain checks, GoPlus Security, and DexScreener liquidity data, and return a strict verdict:

{
  "riskScore": 20,
  "riskLevel": "CAUTION",
  "verdict": "PROCEED WITH CAUTION. Top 10 holders control 51.8% of supply.",
  "shouldExecute": true,
  "reasons": [
    "Top 10 holders control 51.8% of supply."
  ]
}

What Makes This Different

• Deployer wallet freshness — flags newly-created wallets used for one-off token launches
• Brand impersonation detection — flags tokens named after known companies (Apple, Google, Meta, etc.)
• Data source disagreement — flags cases where third-party APIs and our own on-chain checks disagree
• Human-readable verdict — one plain-English sentence, not just raw scores
• Sub-millisecond cached responses — repeat scans within 30 seconds return instantly, with a cached field so you know whether a result is fresh or reused

Quick Testing with MCP Inspector

Want to poke at the MCP server without writing any code? Run: npx @modelcontextprotocol/inspector

Then connect it to https://agentrisk.dev/mcp/manifest and call checktokenrisk directly from the UI.

Cache Freshness Warning

Every response includes cached (boolean) and timestamp (unix seconds) fields. Cached results are served for up to 30 seconds — long enough to speed up repeat lookups, short enough to catch a rug pull or a newly-enabled honeypot function in most cases. For the final safety check immediately before executing a trade, we recommend either calling with a fresh request or checking that cached is false / timestamp is very recent before trusting a shouldExecute: true result.

⚙️ How It Works (M2M Architecture)

1. Agent Discovers Token via mempool or DEX router event.
2. Agent Calls AgentRisk MCP Tool (checktokenrisk) or the direct /scan endpoint.
3. Instant x402 Micropayment ($0.15 USDC instantly settled on Base — no API keys, subscriptions, or credit cards; pure machine-to-machine payment).
4. Binary Decision: The agent receives shouldExecute: true/false with a risk score and structured reasons.

📦 Python SDK

pip install agentriskm2m

import asyncio
from agentrisk import AgentRisk

async def main():
    risk = AgentRisk(private_key="your_base_wallet_private_key")
    result = await risk.scan("0xTokenAddressHere")
    print(result["verdict"])

asyncio.run(main())

PyPI page

Or check a token instantly from the terminal, no code needed:

export AGENTRISK_PRIVATE_KEY=your_base_wallet_private_key
agentriskm2m check 0xTokenAddressHere

🚀 Copy-Paste Integration (60 seconds)

Install the x402 SDK, then run this — it handles payment automatically:

pip install x402 eth-account

import asyncio
from eth_account import Account
from x402 import x402Client
from x402.http.clients import x402HttpxClient
from x402.mechanisms.evm import EthAccountSigner
from x402.mechanisms.evm.exact.register import register_exact_evm_client

Shortened. The full README is on GitHub.