Munin shows how MCP Apps keep human approval inside the chat UI

k_man9 · reddit · 2026-07-29

Munin, an open-source MCP-first system, explains how it uses MCP Apps to keep human approval inside the chat UI while preserving a safety boundary. The agent can draft outbound emails, contact merges, and knowledge-base publishing actions, but irreversible steps still require a person to approve them.

The post describes a practical host-enforced pattern: approve and dismiss are hidden from the model via visibility: ['app'], while the panel runs with the model’s shared credentials and still relies on server-side checks as the real boundary. It also mentions other panels such as asset galleries, analytics funnels, and traffic charts.

Original post →

More from coding & agent

coding & agent channel →