Claude Code's doctor command audits your context budget

Claude Code's /doctor command got rebuilt into a full setup audit in version 2.1.205, per Releasebot.

A typical setup with 5 MCP servers and 58 tools burns about 55,000 tokens before you type a single prompt. That's context the model never gets to spend on your actual problem.

/doctor now flags skills, MCP servers, and plugins you've installed but aren't using. It weighs each one against what it actually costs you in context. It also deduplicates local versus checked-in CLAUDE.md files and calls out hooks that are running slow.

The practical move is two steps. Run /context first to see what's loaded and how much of your window it's eating. Then run /doctor to find what's safe to cut, especially if you inherited a config you didn't build yourself.

Most people running Claude Code have never checked what their own setup costs them. A 55,000-token tax before the first prompt won't stay rare, it'll become the default for anyone stacking MCP servers without auditing them.

The release notes don't say whether /doctor removes flagged items automatically or only surfaces them for you to cut by hand. They also don't say how it weighs a rarely-used server that's still critical against one that's genuinely dead.