AI coding tools like Claude Code can dramatically speed up development, but costs can grow quickly if context and token usage aren’t managed—especially if you don’t use locally running models. Most token waste comes from long conversations, large logs, and unnecessary files being loaded into context.
I’ve collated a short checklist, from one-time setup to daily workflow habits, to reduce token usage, lower cost, and keep Claude responses fast and reliable. These can save 30–50% of your Claude Code cost.
Checklist: Setup and Workflow
| Action | Setup / Example | Why |
|---|---|---|
| Output compression (RTK) | brew install rtk → rtk gain → rtk init --global. Ref: rtk-ai/rtk | Free CLI proxy that compresses outputs for Claude Code, Cursor, or Aider; cuts tokens in console/CLI or VS Code (not web). |
| Language server | e.g. npm install -g pyright or npm install -g typescript-language-server, then in Claude open /plugin and select it | Gives Claude type info, definitions, and references without reading large files. |
| ccusage | npx ccusage@latest → npx ccusage daily --project myapp --breakdown | Shows where tokens are spent by project, model, and time period. |
| Claude monitor | pip install claude-monitor → claude-monitor --plan pro --refresh-rate 5 | Live visibility into token usage while you work. |
.claudeignore | Add node_modules/, logs/, dist/, build/, coverage/ | Stops Claude from scanning large generated folders. |
CLAUDE.md + skills | Keep CLAUDE.md small (wc -l CLAUDE.md); move large workflows from .claude/rules/ to .claude/skills/ | Reduces always-loaded context; large instructions become on-demand. |
| Quick reference file | Create docs/QUICK_REF.md with commands like npm run dev, npm run test, npm run build | Lets Claude use a small cheat sheet instead of large docs. |
| Hooks | .claude/hooks/session-start.sh → git branch --show-current; .claude/hooks/pre-tool.sh → grep "ERROR" logs/app.log | Injects useful context and filters noisy logs before they reach Claude. |
| Session management | /clear, /compact, /rename <name>, /resume <name>, /cost, /model haiku, /mcp disable <tool> | Keeps context small, reduces cost, avoids carrying stale conversations. |
| Prompt and workflow habits | e.g. “Fix race in src/auth.ts lines 42–60”; “Done when: tests pass, no lint errors”; use subagents for multi-file work; use Shift + Tab before implementation | Reduces unnecessary reading, follow-ups, and large-context investigations. |
| Scripts for repeated tasks | format.sh → npm run lint && npm run format && npm run typecheck | Moves repeated work out of chat and into local automation. |
The Main Causes of Token Waste
| Source | Typical Fix |
|---|---|
| Large CLI outputs | Use RTK compression |
| Large generated files | Configure .claudeignore |
| Long conversations | Use /clear and /compact regularly |
Optimising these three areas alone typically reduces token usage by 30–50% while making Claude responses faster and more reliable.