Use /context to Inspect Token Usage
When a Claude Code session starts feeling slow or unfocused, inspect the context before guessing.
Run:
/contextUse it to see what is taking space: messages, files, tools, memory, MCP servers, and other loaded context.
When to use it
Section titled “When to use it”- Before a long refactor.
- After a large investigation.
- When Claude seems to forget earlier instructions.
- Before deciding whether to
/compactor/clear. - When tuning
CLAUDE.md, skills, MCPs, or status line context.
What to do next
Section titled “What to do next”- If the session contains unrelated work, use
/clear. - If the session is still one coherent task, use
/compact. - If always-loaded instructions are too heavy, prune
CLAUDE.md, or let/doctorpropose trims. - If external tools are loading too much, check the tool search setting before removing servers.
MCP servers cost less than they used to
Section titled “MCP servers cost less than they used to”Older advice said to remove MCP servers you rarely use, because every schema loaded into context at startup. That is no longer the default. With tool search on, Claude Code lists only the tool names at startup (about 120 tokens in the docs’ own simulation) and pulls full schemas in on demand.
/context shows what MCP tools currently cost. If the number is large, check ENABLE_TOOL_SEARCH: auto loads schemas up front when they fit in 10% of the window, and false loads everything. Leave it at the default unless you have a reason not to.