Let coding agents update fraud detection rules via MCP

Kairos exposes an MCP server so a user's agent (Claude Code, Cursor, Codex) can maintain fraud decisioning logic as fraudsters change tactics, for fraud teams. · Nov 3, 2026

What They Did

In the Kairos console you turn on the MCP tool groups your agent needs and create a key. You then add the server to Claude Code, Cursor or Codex. The agent builds the workflow and proposes model variants. Kairos trains and tests them on your labeled data, and the agent explains its pick and publishes the winner as the next workflow version.

How It Went

The README's example has an agent design six variants for a Shopify order workflow. It publishes logistic regression plus Grayson answers as v6, with ROC-AUC 0.949 on validation and 0.930 on test. This is an illustrative example, not a reported deployment.

Worth Knowing

Experiments need labeled history, such as chargebacks, to train and test on. Each agent action is recorded under its key, and you choose which tool groups to enable.