I run Claude Code on my own server and drive it from my phone. The 3 things that broke, and what fixed them
retained text
Been doing most of my Claude Code work from my phone for a few weeks, against a session running on my own Linux box. Three things kept biting: 1. The login expires and nothing tells you. A long task dies overnight and you find out in the morning. Fix: catch the signed-out error and show a sign-in step right in the chat on the phone, so it's one tap. 2. Background work finishes while you're away. Claude starts a run\_in\_background job or a Monitor, you close the app, and the result lands in a session nobody is watching. Fix: keep the session warm while background work is running, and when it finishes, wake the conversation with a "background command finished" line, like the terminal does. 3. Permission prompts get lost. The CLI waits on an approval you can't see. Fix: each prompt becomes a card with Allow / Allow for session / Deny, plus a push notification. Curious how others handle the login one in particular. Does anything keep it alive on a headless box, or do you just re-auth every few weeks?
document fetched 2026-11-15 00:22 · canonical https://reddit.com/r/ClaudeAI/comments/1xvz5f6/i_run_claude_code_on_my_own_server_and_drive_it