ccwatts
v0.2.0
Published
Estimate the electricity your Claude Code usage consumes.
Maintainers
Readme
ccwatts
See how much electricity your AI coding uses. ccwatts reads the logs Claude Code already keeps on your machine and turns token counts into an electricity estimate, always as a range with sourced assumptions.
npx ccwatts # your last 30 days
npx ccwatts --by project # also: model, day, type
npx ccwatts history # weekly trend, survives log cleanup
npx ccwatts receipt # share line and card
npx ccwatts dashboard # offline HTML dashboard
npx ccwatts pace --demo # work rhythm (opt-in): active hours, past 10pm, days off
npx ccwatts --method # every parameter and its source
npx ccwatts --demo # no Claude Code logs yet? try synthetic sample data
npx ccwatts help # every command and optionLet Claude Code ask it directly, as a local MCP server on stdio (no port, no network):
claude mcp add ccwatts -- npx -y ccwatts mcpThen ask "how much energy did my sessions use this week?". Tools: energy_summary, energy_by_project, energy_weekly, methodology.
- Local only. No network calls, no accounts, no telemetry. It reads token counts, never prompt or response text.
- Pace is opt-in.
ccwatts pacekeeps~/.ccwatts/pace.jsonl(daily minutes and whole hours only), created only when you run it. Delete it any time. It is never shared unless you passreceipt --share-pace. - Ranges, not false precision. Every number shows a low–high range and the method version.
- No runtime dependencies. Everything, including the MCP SDK, is bundled. Node 22+.
Try it in your browser first, on sample data: https://chaseholroyd.com/ccwatts/
Methodology, source and screenshots: https://github.com/holroydchase-afk/ccwatts
MIT license. The methodology and parameters are CC BY 4.0.
