Passive usage analytics for OpenAI Codex and Claude Code. TokenSight reads the session logs already on your Mac — it spends zero tokens to show you exact totals, cache hit-rates, rate-limit burn-down, and which tools eat your context.
Buy TokenSight — $9.99 Download free trial
Server-reported numbers, bucketed by each request's own timestamp — with per-tool attribution down to the individual MCP call.
Per-project, per-week, per-model token counts straight from the server's own accounting — not estimates.
Which tools, MCP servers and searches caused the most context growth — plus a dominant-consumer label on every single request.
Cache hit-rate, weekly limit %, reset time and plan — the same snapshot the server reports, tracked over time.
One dashboard for both. Streaming partials deduped, cache fields normalized, thread and sub-agent drill-downs.
Daily rate-limit history, a cloud-usage canary, and an estimate of your absolute weekly token budget.
Menu-bar ring, threshold notifications at 50 / 80 / 95 %, live updates as sessions stream in.
TokenSight parses local files your tools already wrote. No agent integration, no SDK, no background traffic. The single network call it can ever make is the license check you trigger yourself.
Checkout via Polar. Download TokenSight, drag to Applications, launch.
License key arrives with your order. One click to validate — then it works offline.
TokenSight immediately scans your existing Codex and Claude history and keeps counting live.
TokenSight