MartinLoop MCP server

MartinLoop is a helper that watches over your AI coding agent while it works. It sets limits on how much it can spend and how many times it can retry, checks the work before calling it done, and gives you a simple report at the end. It is handy if you let an AI write or fix code and want to stop it from going off the rails.

What this MCP server does

You tell your AI what coding job you want done, like 'fix the tests' or 'add a feature'. Your AI uses the MartinLoop helper to set up a run with a spending cap and a retry limit. The helper then lets your coding agent work inside those limits, runs your chosen checks, and stops if something goes wrong. When it is done, it gives you a clear result: verified, stopped, or needs review, along with a receipt you can look at.

Click to zoom

What you can do with it

• Start a governed coding run with a budget and iteration cap
• Run a verifier command to check if the work passes
• See a summary of the latest run and its outcome
• Check how much money was spent on a run
• Look at past runs and their receipts
• Validate the integrity of a run's stored evidence
• Generate a shareable proof card for a run

What it gives back to you

You get back a clear outcome: VERIFIED, STOPPED, or NEEDS REVIEW. You also see details like how much was spent, how many attempts were made, which checks passed or failed, and what files changed. Sometimes it gives you a shareable proof card or a receipt file you can look at.

Before you start

Who it's for

Developers, team leads, or anyone who uses AI coding agents and wants to keep them from running wild.