Have an agent revise fraud rules through MCP

Give a coding agent access to fraud tooling so it can draft, test, and propose rule changes while you keep final approval.

Done with Claude CodeCursorCodexยท Intermediate

What they did
Kairos exposes an MCP server that can be attached to Claude Code, Cursor, or Codex. In the Kairos console, a user switches on the tool groups their agent needs, creates a key, and then registers the server in the coding agent. From there, the agent assembles the workflow, sketches model options, and hands them to Kairos for training and evaluation on labeled examples. The agent then explains the choice and can publish the selected workflow as the next version.
How it went
In the README example, the agent produced six candidate designs for a Shopify order workflow. It selected logistic regression with Grayson answers as v6, with ROC-AUC of 0.949 on validation and 0.930 on test. That example is only a demonstration, not a documented deployment.
Tips
You need labeled historical cases, such as chargebacks, before the system can train and score alternatives. Limit the enabled tool groups to what the job actually needs, and remember that each action is recorded under the key you created.

Try it yourself with Claude Code

Link [Claude Code/Cursor/Codex] to the [MCP server name or URL] for our fraud workflow. Review the current rules and these new patterns: [describe recent fraud behavior]. Draft better rules or model updates with a brief note for each, then show me the proposed changes. Do not publish, apply, send, delete, or submit anything until I explicitly approve it.

Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.

Read the original on github.com โ†—

Get 3 agent uses like this every morning

With the day's top AI news. Free, and one click to unsubscribe.