Blog · Founder note
Why I built Stet
I've spent about fifteen years building products, most of them in fintech, where "it worked yesterday" is not a small thing. When AI coding agents got good, I used them for everything. I shipped faster than I ever had.
Then the bugs started arriving from places I hadn't touched.
The pattern
I'd ask an agent for one change. It would make it, and also tidy a helper, rename a function and adjust a config value it thought was wrong. Each change looked reasonable. Together they broke a flow that had been stable for months. I'd find out from a user, not a test.
I tried the obvious fixes. Better prompts. A longer rules file. More tests. They all helped a bit. None of them stopped it, because none of them checked what the agent actually did.
What I wanted
I wanted three things that felt basic:
- Tell the agent which files a task may change, and get told when it changes anything else.
- Name the behaviours that must keep working, with a command that proves each one, and run them after every change.
- Have every agent, whether Claude Code, Cursor, Codex or a chat window, read the same rules.
Nothing did all three in a way I could set up in minutes, so I built it. "Stet" is the proofreader's mark for "let it stand": keep what's right. That's the whole idea.
Where it is now
Stet is a free, open-source CLI and MCP server. It's on npm as stet-cli, listed in the official MCP Registry, and there's a GitHub Action for pull requests. It runs locally: no account, and your code never leaves your machine.
I'm building it in public, and the best features so far came from people who tried it and told me what annoyed them. If that's you, I'd love to hear it.
Put guardrails on your AI coding agent
Stet is a free, open-source CLI and MCP server for Claude Code, Cursor, Codex and Copilot. It runs locally: no account, your code never leaves your machine.
npx stet-cli init · Install · Setup guide · GitHub