MCP Server
The MCP server lets AI agents work with your rules. Once your application offers it, you can connect an AI client you already use, such as Claude, and ask it in plain language to find rules, explain what a rule does, write a new rule or a new version, write and run tests, or explain why a particular decision was made in production. The agent works through the same Rule Repository and Audit your colleagues use, with your permissions, and everything it does is recorded.
MCP (Model Context Protocol) is an open standard that AI clients use to connect to other software. You do not need to know anything about it to use it: you add your application's address to your AI client once, sign in, and the agent can then use your rules.
What it is good for:
- Finding things quickly: "Which rules use the customer's credit score?", "What changed in the latest version of this rule?"
- Explaining: "What does this rule do?", "Why was order 123 put on hold?"
- Writing: "Write a rule that flags orders over 10,000 shipped abroad", saved as a draft or a new version for a person to review.
- Testing: "Would this rule approve a customer with these values?", "Run the tests of this rule."
- Measuring: "How often did this rule decline applications last month, by region?"
A Person Stays in Charge
The MCP server is built so that agents prepare work and people decide:
- Nothing an agent saves goes live. Every rule or version it saves waits for a person's approval, and it can never make a version live, take one offline, approve, decline or delete anything.
- You confirm what matters. Before an agent changes an existing rule, it asks you. Before a test calls your application's real methods, you are asked to confirm.
- It acts with your permissions. The agent can do only what you are allowed to do, and less: the limits above apply to it even if your own permissions are wider.
- Everything is recorded. The activity log names both you and the agent for every change it makes.
What It Uses
- Your AI client, such as Claude on the web, on your desktop or on your phone, or an AI coding tool. See Connecting an AI Agent.
- Your application's sign-in. The agent never sees your password; you sign in through your company's usual sign-in page.
- The Rule Repository, for sources, rules, drafts, tests and history, and Audit, for production evaluations. Your application may offer one or both; the agent sees only what is there.