CTRLRun MCP server

CTRLRun is a safety layer that sits between your AI and the actions it wants to take. Before a refund, a deletion or a deploy actually happens, CTRRun checks it against rules you wrote down. It is handy for anyone whose AI can do things that are hard to undo.

What this MCP server does

You ask your AI to do something, like refund a customer. Your AI turns that into a tool call, and the call goes to CTRRun instead of straight to the service. CTRRun reads the details of the call, such as the amount, and compares them to your rules file. Small amounts may go through on their own, bigger ones wait for a human to approve, and anything not covered by your rules is refused. Every request, decision and result is written down as a receipt you can look at later.

Click to zoom

What you can do with it

• Set limits on what your AI may do without asking, like refunds up to a certain amount
• Hold risky actions in a queue until a person approves them
• Block actions that fall outside your rules
• Stop the same action from running twice by accident
• Refuse a retry when the outcome of the first try is unknown
• Keep a written record of every request, decision and result
• Let one person or team pass limited authority to another

What it gives back to you

You get a clear answer for each action: allowed, waiting for approval, or refused. If it is waiting, you get a request id you can take to a human for a yes or no. You can also ask for the list of receipts, which shows what was requested, what was decided and what happened. Refusals are listed too, so nothing quietly disappears.

Before you start

Who it's for

Teams whose AI agents can send, pay, refund, delete, deploy or change permissions, and anyone who wants a human in the loop for the risky ones.