Agentrisk

What agentrisk does

The agentrisk MCP server provides pre-trade risk assessments for token contracts on Base. An agent submits a token address and receives structured output including a numeric risk score, risk level, plain-language verdict, reasons, and a shouldExecute boolean. This lets an automated workflow make a conditional decision before attempting a swap on Uniswap, Aerodrome, or another trading path.

The assessment combines direct on-chain checks, runtime simulation, GoPlus Security information, and DexScreener liquidity data. Additional checks cover deployer wallet freshness, possible brand impersonation, and disagreements between external data sources and on-chain observations. Results are intended for machine use, while the verdict and reasons provide readable context for logs or operator review.

How it works

A client connects to the MCP manifest at https://agentrisk.dev/mcp/manifest, where it can discover the check_token_risk tool. The corresponding tool endpoint is POST https://agentrisk.dev/mcp/tools/check_token_risk. A client can also use the direct endpoint, GET https://agentrisk.dev/scan?token=<CONTRACT_ADDRESS>.

Access uses the x402 payment protocol. A scan costs 0.15 USDC on Base Mainnet and is settled through Coinbase's official CDP facilitator. Requests without valid payment information receive HTTP 402 instructions. The supplied wallet needs USDC for the scan and ETH on Base for gas where applicable.

Responses include cached and timestamp fields. Repeat scans may use a result cached for up to 30 seconds. For a check immediately before trading, inspect those fields and prefer a fresh response, particularly before relying on shouldExecute: true.

Setup and configuration

For MCP clients, add the remote manifest URL to the client's MCP configuration:

{
  "mcpServers": {
    "agentrisk": {
      "url": "https://agentrisk.dev/mcp/manifest"
    }
  }
}

The service can be tested with MCP Inspector by connecting to that manifest and invoking check_token_risk. The README also documents a Python SDK named agentriskm2m, installed with pip install agentriskm2m. Its client accepts a Base wallet private key, and the command-line example reads that key from AGENTRISK_PRIVATE_KEY.

For direct HTTP access, the x402 SDK can sign requests with an EVM account. The documented Python examples use x402, eth-account, and a private key, with the account funded on Base. Keep private keys outside source code and provide only the wallet resources needed for payment and gas.

Tools and capabilities

The documented MCP tool is check_token_risk. It accepts a Base token address and returns:

• A risk score and risk level
• A human-readable verdict
• A shouldExecute decision
• Structured reasons supporting the result
• Cache status and a Unix timestamp

The same scan can be called through the direct /scan API, the Python SDK, a terminal command, or a custom bot using the x402 client libraries. The repository also shows how to wrap the scan for LangChain, Coinbase AgentKit, or a plain Python trading workflow.

Limitations and notes

agentrisk is designed for Base tokens and requires payment for each comprehensive scan. The documented price is 0.15 USDC per call; this is not a free blocklist or an API-key-based service. A cached response can be up to 30 seconds old, so cache metadata matters for time-sensitive trades.

A positive shouldExecute result is an input to an automated decision, not a guarantee that a token is safe or that a trade will succeed. The README specifically recommends checking freshness before the final safety check. The provided material documents the check_token_risk tool but does not list additional MCP tools.