Frisk MCP server

Frisk

Pre-transaction risk screening for autonomous AI agents.

Before your agent pays an x402 seller or calls an unfamiliar tool, ask Frisk whether the counterparty is trustworthy and whether the transaction fits your policy. Frisk returns a verdict — allow, review, or block — with a trust score and human-readable reasons. It is advisory: your agent stays in control of the decision.

import { Client } from "frisk-screen";

const client = new Client(); // lite mode, no key required

const result = await client.screen("0x9a3f1b2c3d4e5f60718293a4b5c6d7e8f9a0bc12", {
  endpoint: "https://api.seller.x402/quote",
  amount: 2.5,
  asset: "USDC",
  policy: { maxPerCall: 5.0 },
});

if (!result.allowed) {
  console.log(result.verdict, result.trustScore, result.reasons);
}

Surfaces

Both SDKs expose the same model: a Client with a screen() call, a lite mode that runs locally with zero dependencies, and an optional hosted mode for reputation history and live threat intelligence.

MCP server

For agents that cannot import a library, and for asking the question interactively, the same checks are available as an MCP server exposing one tool, screen_payment:

{
  "mcpServers": {
    "frisk": {
      "command": "npx",
      "args": ["-y", "frisk-mcp"]
    }
  }
}

No API key and no account: with no configuration it screens entirely on your machine. It is listed in the MCP registry as dev.tryfrisk/frisk.

An MCP tool runs only when a model chooses to call it, so a check the model can skip is a weaker guarantee than the same check on the code path that signs the payment. Where the money actually moves, prefer the SDK. Details in mcp/.

Lite mode vs. hosted

Lite mode catches obvious problems — malformed counterparties, payTo swaps, insecure endpoints, policy violations, and a small seed blocklist — without a network call. The hosted API (https://api.tryfrisk.dev) adds reputation history and continuously updated threat intelligence.

What a verdict covers

Shortened. The full README is on GitHub.