Buying snapshot
Scroll sideways to see all columns. With a keyboard, focus this table and use the arrow keys.
| Decision point | What the project states |
|---|---|
License | Open source under the MIT license; you can build and use it yourself for free. |
Where data stays | No cloud, no telemetry, no accounts; everything stays on your machine. |
How you run it | As a native macOS app (bought or built yourself), or in browser mode on any OS, where you run the collector with Node.js and serve the dashboard locally. |
Tools it reads | OpenClaw, Clawdbot, Claude Code CLI, Claude Desktop, Cursor, Windsurf, Cline, Roo Code, Aider, Continue.dev, Codex CLI, Codex Exec and Codex Review. |
How costs are priced | Per-million-token rates: Anthropic's published USD for Claude, OpenAI API standard USD for Codex. |
Our take
Choose Claude Usage Tracker if you are one developer who moves between several Claude or Codex coding tools and wants a private, per-machine estimate of what that work costs. Because everything stays on the machine it runs on, look elsewhere if you need one shared cost view across a team.
Evaluate Claude Usage Tracker
Comparisons
Where it stands out
One view across providers. An All / Claude / Codex toggle scopes the whole dashboard to one provider, and totals stay in one dollar figure. Tools that are not installed or have no data are skipped silently, so there is nothing to configure per tool.
Several ways to find the spend. Daily, weekly, monthly and all-time totals; breakdowns by provider, model and source; cost per project (working directory) with filters for source, model, date range and minimum cost; a peak-hours heatmap; monthly cost projections; and a most-expensive-session callout.
Session detail you can act on. Each session shows its token breakdown (input, output, cache read, cache write, and Codex reasoning tokens), a conversation preview and a one-click resume command (
claude --resume <id>orcodex resume <id>).Pricing rules beyond a flat rate. Claude cache writes are split by TTL from the breakdown Claude Code records per request; fast mode requests are billed at 2× and US-only inference at 1.1×; requests over 272K input tokens use OpenAI's long-context rates on GPT-6, GPT-5.6, GPT-5.5, GPT-5.4 and GPT-5.4 Pro.
Product screenshot
Trade-offs to weigh
Costs are calculated, not billed. The dollar figures come from token counts in local logs multiplied by the project's own rate table. Some inputs are approximations: tools that do not record a cache TTL are priced at the 5-minute cache-write rate, Codex logs give cached-input tokens but no cache-write tokens, and Codex models that OpenAI no longer lists use their last published rates. Treat the totals as estimates and check your provider invoice for the actual bill.
Rates need to keep up. Accuracy depends on the rate table being current; the README dates its check to 2026-10-14 and notes that GPT-5.6 Sol is on promotional pricing through at least 2026-11-21.
macOS first, fixed tool list. The signed app needs macOS 12.0+, and other systems use browser mode. The README lists Linux / Windows path support, Gemini CLI and Copilot CLI support, CSV/JSON data export and cost alerts or budget thresholds under "Ideas for contributions", so check the current release if you need any of them.
Pricing and plan conditions
What you pay for is convenience: the README offers the Premium Build to people who want a ready-to-use macOS app instead of building one, and calls it the easiest way to use the app on macOS. On other operating systems, browser mode is the route.
Scroll sideways to see all columns. With a keyboard, focus this table and use the arrow keys.
| Plan | Published price | What it adds |
|---|---|---|
Build from source | Free | Clone the repository and run ./build-app.sh; open source, without the premium license gate or premium update checker. |
Premium Build | $9 one-time purchase | Signed macOS app, license management, automatic update checks and one-click updates; installed from a .dmg. Requires Node.js v16+ and macOS 12.0+. |
Public signals
About these signals
Verified factual signals from public sources. They indicate observable activity or interest, not total adoption, product quality, or cost.
18 GitHub commits 90d61 GitHub stars
See all signals from 3 sources
Alternatives to Claude Usage Tracker
The reviewed substitutes for Claude Usage Tracker among the claude code tooling, and what would make each one the better answer.
Direct alternatives
Reviewed substitutes: products bought for the same job, where a team picks one.
- CCDash
- Both read Claude session data off the local disk and show what it cost, and a developer installs one of them. The 2026-09-10 rejection under IDENTITY-007 read CCDash as a session monitor against Claude Usage Tracker's cost tracker -- two different jobs. That was true of CCDash's March launch and is no longer true of the product: github.com/zihenghe04/CCDash now leads with 'track tokens, quota, costs across Claude Code, claude.ai and the API in one panel' and has dropped task scheduling. The split is breadth against depth. Claude Usage Tracker reads 10+ tools -- Cursor, Windsurf, Cline, Aider, Continue.dev, Codex -- and reports historical cost. CCDash reads only Claude surfaces but adds what the other does not have: 5-hour and 7-day subscription quota tracking, burn-rate and rate-limit prediction, budget thresholds and Slack or Discord alerts.Applies to: Seeing what AI coding assistants cost, from local session files, without an API key. Claude Usage Tracker stands in when the spend is spread across several editors and agents and the question is what was spent. CCDash stands in when the work is Claude-only and the question is how close the next throttle is.
Other approaches
A different approach to the same problem. Each substitutes only for the workload named beside it.
- Auditi
- Both answer 'what are my AI tools doing', measuring different things: Claude Usage Tracker attributes cost across OpenClaw, Claude Code, Claude Desktop and Cursor, while Auditi watches agent performance and behaviour. A team adding visibility picks the measurement that matters first.Applies to: Whether AI agent visibility should start with cost attribution or with behaviour.
Related technologies
Normally used together rather than chosen between, so these are not alternatives.
- capsul
- Both address the cost of Claude Code and Codex, from opposite ends. Claude Usage Tracker reads local session files and reports what was spent across tools; capsul sends less context per request and keeps its own local ledger. A developer can run both: one shows where tokens went, the other lowers what each request sends. Neither replaces the other.Applies to: Cutting Claude Code or Codex spend: measure with the tracker, lower per-request context with capsul.