silverdolphin863/claude-code-meter
Tray widget showing Claude Code and Codex rate-limit usage and burn pace. Windows, local-only, MIT. - silverdolphin863/claude-code-meter
0 stars · 0 forks · 0 watchers
README
CC Meter A tray widget for Windows that shows how much of your Claude Code and Codex rate limits you have burned, and how fast you are burning them.
Representative usage rendered by the v1.0.4 full panel at native resolution. Not a cost tracker. It answers the only question that matters mid-session: am I going to hit the wall before this window resets? What it shows For each limit window: percent used, a bar, the reset time, and a pace badge. Pace is the useful part. 1.0x means you are burning quota exactly in step with the clock, so you will run out precisely at the reset. 0.3x means you are coasting. 1.6x means you will hit the wall early, and roughly how early.
Claude Code - 5-hour window, weekly all-models, and weekly per-model (Opus, Fable, whichever your plan scopes). The per-model split is real, it comes from the same source the /usage screen uses. Codex - the account-wide windows Codex currently reports. One shared pool; some plans expose only a weekly window, and CC Meter never invents a missing one.
Two modes: a compact strip that docks against the top edge of the screen (with optional auto-hide until you touch the edge), and a full panel with countdowns.
The compact mode stays out of the way at the top edge. Open the compact strip at its full native resolution.
Install Download CCMeter-Setup-x.y.z.exe from Releases and run it. It lives in the tray, not the taskbar. The installer is unsigned, so Windows SmartScreen will show "Windows protected your PC". Click More info -> Run anyway. If you would rather not trust a binary from a stranger on the internet, build it yourself from the source below. That is the honest answer and it is why the source is here. Where the numbers come from CC Meter has no hosted service. It makes the Claude usage request to Anthropic from your machine with your own OAuth token. For Codex, it asks the installed Codex CLI for the current account snapshot; Codex handles its own authenticated OpenAI request. CC Meter never reads the Codex authentication token. Claude: reads the standard Claude Code OAuth token from ~/.claude/.credentials.json and calls https://api.anthropic.com/api/oauth/usage directly when the local cache is older than ten minutes. It writes the response to ~/.claude/usage-cache.json and honours the cache lock. No status-line script is required. Codex: launches the local Codex app-server and calls account/rateLimits/read, cached for five minutes. This returns the current account-wide windows even when this machine's session files are old. If an older Codex version does not expose that method, CC Meter falls back to recent files under ~/.codex/sessions/ and discards every window whose reset time has already passed. About that token CC Meter reads your Claude OAuth token off disk and sends it to api.anthropic.com and nowhere else. It is never logged, never stored anywhere new, and never transmitted to any host but Anthropic's. The direct Claude network path is isolated in refreshClaudeUsage() in server.mjs. Codex authentication and account requests remain inside the installed Codex CLI. The usage endpoint is undocumented and rate-limited (observed: HTTP 429 with a ~57 minute Retry-After). CC Meter refreshes at most once every 10 minutes, honours Retry-After exactly, backs off exponentially on failure, and persists that backoff across restarts so relaunching cannot hammer a locked-out endpoint. Being undocumented, the endpoint may change or stop working without notice. Local HTTP server The UI is served from http://localhost:7373, and GET /usage.json returns the local usage data. Browser cross-origin access is disabled by default. If you want to drive another display, set CCMETER_CORS_ORIGIN to that display's specific origin before starting the server. No token or session content is returned by this endpoint. Build it yourself npm install npm test # isolated Claude and Codex fixture test npm start # dev npm run dist # unsigned NSIS installer -> dist/ Requires Node 20+ and Windows. macOS and Linux are untested; the tray behaviour and the screen-edge dock are Windows-shaped. Limitations
Windows only, for now. Only works with an OAuth (subscription) Claude Code login. API-key users have no plan limits to display. Live Codex readings require a Codex CLI version that exposes account/rateLimits/read. Older versions use the filtered session-log fallback and update only when Codex writes a session file. No auto-update. Watch the releases page.
License MIT. See LICENSE. Not affiliated with, endorsed by, or supported by Anthropic or OpenAI.