See where your Claude Code tokens go.
tokometer reads Claude Code's own session transcripts and turns them into a live dashboard — real token accounting, full history, and a per-prompt cost breakdown. No telemetry exporter, no env vars, no Prometheus, no Grafana, no Docker.
The whole picture, with nothing to wire up.
Every assistant message Claude Code writes already carries the API response's real
token usage, the model, and a timestamp. tokometer just reads it.
Zero setup
No OTEL_* env vars, no exporter, no restarting your session. Run the binary and open the dashboard.
Real token accounting
input / cache-read / cache-creation / output come straight from each request's usage — exact, not estimated.
Full history, retroactively
Every past session counts the moment you run it — not just the ones after you turned something on.
Per-prompt breakdown
Which prompts cost the most and why, with the real prompt text and the tools each one triggered.
Context attribution
Which files and tools fill the window, so you can see what's making every turn expensive.
Live tailing
Tails the transcripts and pushes updates over WebSocket — new activity shows up within a couple of seconds.
Read the files. Skip the pipeline.
Claude Code writes every session to a JSONL transcript under ~/.claude/projects.
tokometer reads all of it once, then tails each file for new lines.
Three ways in. No Go required.
Prebuilt binaries for macOS and Linux (amd64 + arm64). Then just run tokometer.
| Requirement | Notes |
|---|---|
| Claude Code | tokometer reads the transcripts it already writes to ~/.claude/projects. Nothing to enable. |
| macOS or Linux | Prebuilt amd64 and arm64 binaries. Windows: grab the .zip from Releases. |
| A free port | Serves on :4318 by default — override with -addr. |