Claude Code token usage optimization strategies for developers

Tokenizer explains token concept, claude Code uses a tokenizer to split text into tokens like words, punctuation, and spaces.

Tokens distinguishes input and output tokens, input tokens include all text sent to the model and hidden context; output tokens are text generated by Claude.

Context components counts context components as input tokens, conversation history, CLAUDE.md, tool definitions, and referenced files are tokenized and counted as input tokens.

Token usage tools provides methods to check token usage, use /context command, inspect JSONL logs in ~/.claude/projects/, or use web token calculators.

CLAUDE.md recommends minimizing CLAUDE.md size, shortening CLAUDE.md (example: from ~5,000 to ~700 tokens) reduces initial context cost.