Show HN draft page

Show HN: I made an MCP that tells agents how not to build empty-shell automation.

The failure pattern I kept seeing: people ask an agent to automate work, and it creates files, buttons, or a dashboard that looks finished but does not run the real workflow. This MCP sits before execution and returns the missing pieces: tool route, copy-paste command, dry-run, QA checks, recovery path, and human boundary.

Try it server.json Source

What I want feedback on

Is this useful as an MCP tool, or should it be a static checklist?
Which automation domains should the router cover first: docs, browser ops, coding, research, or knowledge bases?
What would make the output concrete enough to use before running Codex, Claude, ChatGPT, or Cursor?

Example input

Automate competitor research from websites, turn it into an Obsidian learning note, and produce a client-ready report.