Updated
Updated · Android Authority · Jun 9
Claude Audits Home Assistant, Flags 3 Duplicate TV Entries and Builds New Automations
Updated
Updated · Android Authority · Jun 9

Claude Audits Home Assistant, Flags 3 Duplicate TV Entries and Builds New Automations

1 articles · Updated · Android Authority · Jun 9

Summary

  • A local Claude setup parsed the author’s Home Assistant instance, exposed years of clutter, and then built and debugged a new study-lighting automation in minutes.
  • Using an MCP proxy and a long-lived access token, the AI stayed on the local network while gaining read and control access to devices, climate settings, media players, timers, and notifications.
  • The audit found concrete redundancies: one TV appeared under 3 entities, an LG webOS set had 2 unavailable duplicates, and a living-room streamer showed up 4 times alongside ghost components.
  • When Home Assistant’s sandbox blocked direct file writes, Claude used browser automation to enter YAML through the web UI, then traced a logic bug that kept a bookshelf light from turning off.
  • The experiment also highlighted limits: browser-driven edits were slow and fragile, and giving an AI JavaScript-capable access to locks, cameras, or climate controls raises clear security and misconfiguration risks.

Insights

Can AI fix our messy smart homes without introducing catastrophic new risks like hallucinations and system compromise?
When AI manages your lights and locks, who is truly in control—you, the AI, or its developer?