Claude Usage Tracker

What is Claude Usage Tracker?

Claude Usage Tracker is a local-first tool that automatically discovers and aggregates Claude AI usage costs across multiple local development tools. It scans known data directories, parses JSONL/log files, calculates costs using model-specific pricing, and displays the results in an interactive dashboard.

The core purpose is cost tracking and visualization—so you can review spending by source, model, and time period without sending usage data to the cloud. Everything runs on your machine and the dashboard is powered by Chart.js.

Key Features

• Auto-detects 9+ Claude-integrated tools and merges usage data across them for a consolidated view.
• Cost calculation using model-specific per-million-token pricing for Anthropic model families (Opus, Sonnet, Haiku), including input/output and cache read/write where applicable.
• Dark-themed interactive dashboard using Chart.js with visual charts and breakdowns.
• Time-based cost tracking with daily, weekly, monthly, and all-time cost views.
• Model and source analytics including per-model cost breakdowns and filtering (source, model, date range, and min cost).
• Peak pattern visualization via heatmaps and two views: Peak Hours (day × hour grid) and Peak Days (GitHub-style calendar).
• Session-level details including an expandable session log with color-coded source cards, per-session cost callouts (e.g., most expensive session), and a session detail panel.
• Project cadence projection with monthly projections based on the current spending pace.
• macOS app bundle option to build a standalone .app for double-click launching.

How to Use Claude Usage Tracker

1. Install and launch either the latest release (recommended) or run the tool locally from source.
2. Run a scan to collect data and render the dashboard: the macOS app collects fresh data and displays the interactive dashboard, and the browser mode starts a local server and loads dashboard.html.
3. Review the dashboard to explore total costs and break them down by source/model and by time periods using the available charts, heatmaps, and filters.

Quick start options (from the repository):

• macOS (Apple Silicon/Intel): download the latest release, unzip, move Claude Usage Dashboard.app to Applications, then double-click to launch.
• Build from source: clone the repo, run ./build-app.sh, then double-click the built app.
• Browser mode (any OS): run node collect-usage.js, start a local server with python3 -m http.server 8765, and open http://localhost:8765/dashboard.html.

Use Cases

• Consolidate spending across multiple Claude tools: If you use several local integrations (for example, Cursor, Windsurf, Claude Desktop, and Continue.dev), the tracker merges usage from these sources into one dashboard.
• Spot when spend is highest: Use Peak Hours (day × hour heatmap) and Peak Days calendar to identify the times and days with the most expensive usage sessions.
• Audit specific sessions or days: Expand the session log for a date to inspect per-session costs and see token breakdown and conversation history in the session detail panel.
• Compare day-over-day changes: Use Yesterday Delta to compare today’s spending against yesterday at a glance.
• Plan expectations for the current month: Review Monthly Projections to estimate projected monthly costs based on current spending pace.

FAQ

• Does the tracker send usage data to a server? No. The project description states there is “No cloud” and “No telemetry,” and that everything stays on your machine.

• Which tools does it support? The repository lists Claude usage from OpenClaw/Clawdbot, Claude Code CLI, Claude Desktop (local agent mode), Cursor, Windsurf, Cline, Roo Code, Aider, and Continue.dev. Tool detection is automatic.

• What data formats does it read? It scans known directories and parses JSONL/log files; the supported tools section specifies JSONL for each listed tool.

• What if a tool isn’t installed or has no usage data? The tool detection behavior is described as: if a tool isn’t installed or has no data, it is silently skipped.

• How does it calculate costs? Costs are calculated using Anthropic’s per-million-token pricing for the supported model families (Opus, Sonnet, Haiku), based on input/output and cache read/write where provided.

Alternatives

• Local log analysis + spreadsheets: Export or collect your Claude usage logs and compute costs manually in a spreadsheet. This is more manual than an interactive dashboard, but it can fit custom workflows.
• Vendor-provided usage dashboards (if available): Some providers offer usage views for billing/usage reporting. Those typically focus on a single product/account rather than aggregating across multiple local tools.
• General developer time/cost analytics tools: Tools that analyze IDE activity or compute costs based on usage events can help, but may not model Claude-specific token pricing or automatically parse each tool’s storage format.
• Other local-first monitoring dashboards: Adjacent tools that visualize local logs in a browser can provide similar charts, though they may require you to prepare data rather than relying on the tracker’s auto-detection and JSONL parsing.