Auto-discovered marketplace from stuartshields/claude-co2-status-line
npx claudepluginhub stuartshields/claude-co2-status-lineStatus line showing cumulative token usage, energy estimate, CO2 equivalent, and water usage for the current session
A Claude Code plugin that shows how much energy, CO2, and water your session is burning through.
⚡ 40.0 Wh │ 🌱 21.9g CO2 │ 💧 137ml │ 📊 100.5k tokens
After each interaction, the status line shows four things:
Numbers accumulate across the session and update after each assistant message.
/plugin marketplace add stuartshields/claude-co2-status-line
/plugin install co2-status-line@claude-co2-status-line
Then run the setup skill to configure your statusline:
/co2-status-line:setup
The setup skill detects any existing statusline and offers to wrap it (keeping your current statusline on line 1, CO2 metrics on line 2).
/co2-status-line:setup handles this automatically — it detects your existing statusline (GSD, claude-hud, etc.) and offers to wrap it using the --wrap flag. The wrapped command's output shows on line 1, CO2 metrics on line 2. If the wrapped command fails, you still get the CO2 line.
Add --track to the statusline command in ~/.claude/settings.json after setup:
{
"statusLine": {
"type": "command",
"command": "node \"<plugin-path>/src/statusline.js\" --track"
}
}
Totals get saved to ~/.claude/co2-totals.json and show up as a second line:
⚡ 25.4 Wh │ 🌱 13.9g CO2 │ 💧 87ml │ 📊 53.0k tokens
∑ 1.2 kWh │ 687.4g CO2 │ 4.1L │ 2.4M tokens │ 14 sessions
Without --track, it's session-only and nothing hits disk.
Run /co2-status-line:update inside Claude Code.
/plugin uninstall co2-status-line@claude-co2-status-line
Then remove the statusLine entry from ~/.claude/settings.json.
Clone the repo and load it directly:
claude --plugin-dir ./claude-co2-status-line
Or point your statusline at the source:
{
"statusLine": {
"type": "command",
"command": "node \"/path/to/claude-co2-status-line/src/statusline.js\""
}
}
These are napkin math estimates, not precise measurements. Frontier labs don't publish per-token energy data, so the figures come from publicly available research.
Energy per token comes from Epoch AI's analysis of ChatGPT energy usage (Josh You, 2025), adapted for Claude Code by Simon P. Couch in "Estimating the Environmental Impact of Claude Code".
Here's how the Epoch AI methodology works:
input_rate = Total_Wh / (input_tokens + 5 x output_tokens).| Token type | Wh per million tokens | Where it comes from |
|---|---|---|
| Input | 390 | Base rate, back-calculated from Epoch AI |
| Output | 1,950 | 5x input (API pricing ratio) |
| Cache creation | 490 | 1.25x input (cache write pricing) |
| Cache read | 39 | 0.1x input (cache read pricing) |
CO2 per kWh uses 548 g CO2/kWh, the US data center average from a 2024 ScienceDirect study of 1,795 data centers. That's a lot higher than the national grid average of 367 g/kWh (US EIA, 2023) because data centers tend to sit in regions with dirtier power. About 56% of US data center electricity came from fossil fuels in 2023-2024.
Water per kWh uses constants from "How Hungry is AI?" (2025), a paper that benchmarked 30 LLMs including Claude 3.7 Sonnet. Water consumption has two parts: on-site cooling (0.30 L/kWh) and off-site water used to generate the electricity (3.142 L/kWh, US average). The electricity generation side is about 92% of the total.