@tokz/cli
See what your coding agents spend
Your coding agents write every token they burn to a log, then forget about it. tokz reads those logs and shows you where the money went.
$
npx @tokz/cliFree, MIT, and local by default — it reads logs already on your disk and uploads nothing.
- Claude Code
- Codex
- OpenCode
- Gemini CLI
- Qwen Code
- Droid
- Copilot CLI
- Goose
- Amp
- + 8 more
$1,403.90one real 32-day audit
- Per-MCP-server accounting
- Which servers you are paying for, and which ones sat in your context all month without being called.
- Cache savings, separated out
- Cached reads bill at a fraction of list. Counting them as full-price tokens overstates a bill; tokz reports them apart.
What the terminal UI actually shows
tokz — overview
| Model | Tool | Calls | Tokens | Cost |
|---|---|---|---|---|
| claude-sonnet-5 | Bash | 412 | 1.8M | $21.60 |
| claude-sonnet-5 | Read | 968 | 3.1M | $37.20 |
| gpt-5.2 | web_search | 51 | 0.4M | $6.80 |
| claude-opus-5 | Edit | 133 | 0.9M | $54.00 |
cached reads shown apart from list-price tokens, not blended into them
Illustrative — run the CLI on your own logs for your own numbers.
Once you can see the bill, the next question is what to cut
The CLI tells you where the tokens went. The hosted API removes them — same verbatim guarantee, applied to your agent's tool output as it is sent, not to a log after the fact.