Kepil MCP server

Kepil is a helper program that keeps your AI agents on a short leash. Every agent gets an identity card, every job gets written rules, and every action it takes is checked and written down. It is handy if you let an AI do real work, like sending messages or handling documents, and you want to know exactly what it did and be able to undo it.

What this MCP server does

You ask your AI to do a job, like send a message or process a document. The AI uses this helper to check with Kepil whether that action is allowed for this agent and this job. Kepil looks at the rules you set, like spending limits and which systems are allowed. If the action is risky or not allowed, Kepil refuses it or asks a human to confirm. Every allowed action is written into a log that cannot be quietly changed later.

Click to zoom

What you can do with it

• Create a job order with rules and limits for your AI
• Run one step of a job and see if it is allowed
• Check the status of a job and what has happened so far
• See which actions are waiting for a human to approve
• Read an agent's identity card and its limits
• Verify that the action log has not been changed
• List the kinds of jobs Kepil knows how to handle

Try asking your AI

• “Create an order for the inbound leads job with a spending limit of 50 dollars”
• “Run the next step of order ord-0042”
• “What is waiting for my approval right now”
• “Verify the journal and tell me if anything was changed”

What it gives back to you

You get clear answers in the chat. Kepil tells you if an action was allowed, refused, or needs a human to confirm. It can show you a list of pending confirmations, the status of a job, or the result of checking the log. If the log has been tampered with, it tells you exactly where.

Before you start

Who it's for

Anyone who lets an AI agent do real tasks and wants to keep control, like operations managers, small business owners, or team leads who use automations.