cc-monitor 0.1.14

Real-time token usage and cost monitor for Claude Code sessions, in your terminal.

cc-monitor

Real-time token usage and cost monitor for Claude Code sessions, in your terminal.

A Textual TUI that tails Claude Code's session logs, correlates them with the hook event stream, and gives you a live view of every active project, every session, and every model — with live 5h-block tracking so you know how close you are to your plan limit before the bill (or the rate limit) catches you off guard.

---

Table of contents

• Features
• Screenshots
• Installation
  • uv (recommended)
  • pipx
  • pip
  • From source
• Quick start
• CLI flags
• Configuration
• Development
• Acknowledgements
• License

---

Features

• Live 5h-block tracking — progress bar against your Anthropic plan limit (auto-derived P90 in offline mode; authoritative /api/oauth/usage numbers when OAuth is available)
• Toast notifications at 80% and 100% of cost ceiling so you don't have to babysit the screen
• Sessions / Projects / Models tabs with click-to-sort columns, date / cost / model filters, and full keyboard navigation
• Per-session detail screen — usage charts, tool counts, file-read / file-write hot lists, full per-turn context window utilization
• Project detail screen — same metrics rolled up across every session in the project, plus 7-day cumulative cost / token charts
• CSV / JSON export of sessions, projects, and models for pandas / Excel analysis
• Cross-run state snapshot — quit and relaunch in <1s even with hundreds of sessions and an 8-day archive
• Cache-aware detail screens — first open parses the JSONL once; re-opens are instant
• Mouse and keyboard fully wired — clickable filter chips, back button, sort headers, drill-in rows; or use 1/2/3///,/? for everything

Screenshots

Main view (Sessions tab)	Session detail

Project detail	Settings

Installation

Supported platforms: Linux, macOS, and Windows. Windows works in archive-viewer mode without Claude Code installed (just point cc-monitor at copied ~/.claude/projects/ data); native Claude Code on Windows is supported the same way as POSIX. Please open an issue if you hit platform-specific bugs.

uv (recommended)

uv tool install cc-monitor

uv tool creates an isolated environment automatically, sidesteps PEP 668 ("externally-managed-environment") errors on modern Linux, and adds the entry-point shim to ~/.local/bin without touching system Python.

pipx

pipx install cc-monitor

pip

pip install --user cc-monitor

On Debian/Ubuntu/Fedora-based systems with system Python you'll need --user to bypass PEP 668. Prefer uv or pipx instead.

From source

git clone https://github.com/boldsamurai/cc-monitor.git
cd cc-monitor
uv tool install .

If cc-monitor isn't on PATH after install, add ~/.local/bin to your shell's PATH.

Quick start

Just run:

cc-monitor

On first launch it will:

1. Install a hook into ~/.claude/settings.json so Claude Code feeds it tool_start / tool_end events (idempotent, safe to re-run).
2. Replay every JSONL in ~/.claude/projects/ to populate the 8-day archive.
3. Auto-detect OAuth credentials. If found, pull authoritative 5h / 7d utilization from Anthropic. Otherwise fall back to a P90-derived local ceiling.
4. Drop you into the Sessions tab.

Press ? (or ctrl+h) at any time for the keyboard cheatsheet.

CLI flags

Flag	Default	Notes
--poll N	0.5	Polling interval (seconds) for the JSONL tailer.
--max-5h-cost X	—	Pin a custom USD ceiling for the 5h-block progress bar. Without it, --no-api mode auto-derives a P90 ceiling from your last 8 days.
--no-api	off	Disable Anthropic /api/oauth/usage polling. Stay fully offline; falls back to the local P90 ceiling.
--debug	off	DEBUG-level logging to ~/.cache/cc-monitor/usagemonitor.log (rotates at 10 MB).
--reinstall-hook	—	Re-run the hook installer and exit without launching the TUI. Idempotent — safe in provisioning scripts.
--rescan	—	Discard the cached state snapshot before launch so the next run replays every JSONL from scratch. CLI equivalent of Settings → Force re-scan.
--no-update-check	—	Skip the once-per-24h check against PyPI for a newer cc-monitor release. The check normally runs in the background, never blocks startup, and only shows a toast on launch when an update is available.
--skip-claude-check	—	Skip the startup probe that warns when Claude Code is missing. The probe checks for the claude binary on PATH and a non-empty ~/.claude/projects/, and blocks the main view behind a Continue/Quit modal when both signals are absent. Useful for CI / scripted runs.

Everything else (theme, date format, default tab, refresh interval, filter persistence, confirms) is editable from the in-app Settings screen — , key from the main view.

Configuration

Path	Purpose
~/.config/cc-monitor/config.json	Settings persisted from the in-app Settings screen.
~/.cache/cc-monitor/state.pickle	Cross-run state snapshot. Missing / corrupt / version-mismatched → fall back to full replay.
~/.cache/cc-monitor/usagemonitor.log	Rolling log file (10 MB cap).
~/.cache/cc-monitor/exports/	Timestamped CSV / JSON dumps from Settings → Export.

None of the above is committed to your repo.

Development

git clone https://github.com/boldsamurai/cc-monitor.git
cd cc-monitor
uv sync                      # install runtime + dev deps
uv run pytest                # 88 tests
uv run cc-monitor       # launch from source

Acknowledgements

Built with Textual and plotext (via textual-plotext). The 5h-block detection algorithm is informed by Maciek-roboblog/Claude-Code-Usage-Monitor.

License

MIT — see LICENSE.