Claude Code Context Window Cheat Sheet

The token costs, commands, and decision frameworks from 12+ months of daily Claude Code use, distilled into a single two-sided reference card. Print it, pin it, or keep it open in a tab while you work. Stop guessing where your tokens are going. Start every session knowing exactly what you have to work with.

What's inside

• Context window anatomy — visual breakdown of where your 200k tokens actually go (system prompt, tools, MCP servers, agents, memory, autocompact buffer)
• MCP server token costs — real measurements for Sentry (~19k), Memory (~5.6k), Context7 (~1.8k), Playwright (~14.3k), and the rule of thumb for estimating any tool
• Decision flowchart — when to use a subagent vs. skill vs. MCP server vs. custom agent vs. inline work
• When to /clear vs. /compact — the decision tree, including the warning signs that you need a full reset
• CLAUDE.md optimization checklist — five items to audit so your project config isn't burning tokens on stale instructions
• The "lost in the middle" rule — what it means for your CLAUDE.md structure and long conversations