🚨 URGENT: Claude Code is silently burning my session usage. Has anyone found a fix?

I’ve got a client call today and I’m effectively blocked because Claude Code is unusable right now.

I’m on Max 5x. As you can see in the attached GIF, my current session usage was at 49% and a slow silent consumption keeps going on. So the issue doesn’t look like normal weekly exhaustion. It looks like something inside the current Claude Code session is burning context extremely fast.

I’ve disconnected MCP servers, tried reducing background activity and waited for my weekly limits to reset, but the same issue keeps coming back. Even when I’m barely doing anything, the session usage climbs very quickly.

I’m trying to work out whether this is caused by bloated claude.md files, duplicated project context, hidden MCP activity, background indexing, prompt injection or something else repeatedly loading into context.

Has anyone built a practical way to audit what Claude Code is actually consuming?

I’m not looking for generic advice like “wait for your limits to reset” or “upgrade the plan”. I need to understand what is consuming the tokens so I can fix the setup and get back to client work.

Any transparent breakdown, diagnostic approach or tool/plugin idea would be really appreciated.

19 comments