See what is eating your AI's memory with /context

Your AI gets slower and starts forgetting things when its memory, the context window, fills up. Here is how to look inside it.

See it

/context

It draws your context window as a coloured grid, shows what takes the most room, and gives tips for the heavy parts.

Tools can use space before you type

Connected tools (MCP servers) can take up room before you type a single word. Switch off the ones you are not using:

/mcp

Running out of room? Compact

/compact keep only the auth work

/compact summarises the chat. Tell it what to keep, and it focuses the summary on that.

Between unrelated jobs, clear

/clear

Old chat is sent again with every message, so stale context costs tokens each time. /clear costs nothing and starts fresh.

Rule of thumb

Look with /context before it overflows, compact when a job is long, clear when the next job is unrelated.

Signs it is time to look

• Claude asks something you already answered.
• It forgets a rule you set at the start of the session.
• Replies get slower in a long session.

When you notice these, run /context first to see what is taking the room, then compact or clear using the rule of thumb above.

More quick tips

• Get a report on how you use Claude Code
• Ask Claude Code a side question without breaking its focus
• Customize Claude Code's status bar by describing it
• All terminal tips and tricks