Token Monitor — AI Context Tracker

Real-time context & quota monitor for Claude.ai — see usage live and get smart alerts before your replies are cut off.

Token Monitor gives Claude.ai users real-time, at-a-glance visibility into every dimension of their AI usage — so you never get blindsided by a cut-off reply or an unexpected quota limit. ───────────────────────────────────────────────────────── 🔍 WHAT YOU GET ───────────────────────────────────────────────────────── • CONTEXT WINDOW GAUGE See how full the current conversation is — as a percentage and as a token count. Know exactly when you're approaching the model's memory limit before your next message is sent. • 5-HOUR & WEEKLY QUOTA BARS Claude Pro and Max plans throttle usage over a 5-hour rolling window and a weekly budget. Token Monitor shows both bars in real time and estimates when each will reset. • TRUNCATION RISK WARNING ⚠ Before you hit Send, Token Monitor calculates whether your next message — combined with the predicted output — will overflow the context window. An inline banner warns you with an actionable suggestion: "Split into smaller questions" or "Start a new chat." • OUTPUT SIZE PREDICTION 📊 A small pill next to the Send button predicts whether the reply will be Small, Medium, Large, or XL — so you can plan accordingly. • PER-TURN TOKEN COST Token cost badges appear beneath every user message, showing input and output token counts for that turn. • STREAMING AWARENESS While Claude is generating, Token Monitor tracks tokens already committed to the input and tokens streamed back in real time. • SELF-CALIBRATING ESTIMATES Token counts are estimated via a fast heuristic (no API calls). When Claude shows "X messages left" banners, Token Monitor uses those signals to self-correct its estimates over time — getting more accurate the more you use it. ───────────────────────────────────────────────────────── 🔒 PRIVACY & PERMISSIONS ───────────────────────────────────────────────────────── Token Monitor is built with a privacy-first design: • All processing happens locally in your browser. Your conversations are never transmitted to any external server. • The extension only runs on claude.ai domains. • Quota data is fetched directly from Claude's own settings endpoint (the same data shown in your account dashboard) — no third-party services are involved. • No account creation required. No login. No analytics. Permissions used: storage — saves your preferences and calibration data locally alarms — schedules periodic quota refreshes notifications — optional limit-reached desktop alerts scripting / tabs — required to inject the UI into claude.ai ───────────────────────────────────────────────────────── ⚙️ SUPPORTED PLANS & MODELS ───────────────────────────────────────────────────────── Works with: • Claude Free, Pro, Max (5x and 20x), Team, and Enterprise accounts • All Claude models available on claude.ai: Sonnet, Opus, Haiku • Conversations inside Projects (project knowledge token cost included) ───────────────────────────────────────────────────────── 🌐 LANGUAGES ───────────────────────────────────────────────────────── UI available in English and 简体中文 (Simplified Chinese). ───────────────────────────────────────────────────────── 📌 NOTES ───────────────────────────────────────────────────────── • Anthropic occasionally changes the claude.ai DOM structure. If an indicator stops updating, check for an extension update. • Token estimates use a heuristic (not the exact tokenizer). The self-calibration loop corrects drift over time. • This extension is an independent tool — it is not affiliated with or endorsed by Anthropic.