Claude Counter

Claude Counter adds three lightweight, read-only overlays directly onto claude.ai:

TOKEN COUNT
Shows an approximate token count for your current conversation, plus a mini progress bar scaled to Claude's 200,000-token context limit. Useful for knowing how close you are to a context reset before it happens.

CACHE TIMER
Displays a live countdown showing how long your conversation remains in Claude's prompt cache. While the cache is active, continuing a conversation costs significantly fewer tokens. The timer helps you decide when to reply to get the most out of caching.

SESSION & WEEKLY USAGE BARS
Reads your usage data directly from Claude's own /usage API endpoint and live SSE stream. Displays a 5-hour session bar and a 7-day weekly bar, each with a real-time reset countdown. Because it reads the raw utilization fraction from the SSE stream (rather than the rounded numbers shown on Claude's usage page), the bars are more precise than what Claude natively displays.

PRIVACY & PERMISSIONS
- Runs only on https://claude.ai — it never activates on any other site
- Makes network requests only to claude.ai, using your existing login session
- All token counting and data processing happens locally in your browser
- No data is collected, stored externally, or sent to any third-party server
- No analytics, no tracking, no ads

TECHNICAL NOTES
Token counting uses a locally bundled copy of the open-source gpt-tokenizer (o200k_base). Counts are approximate and exclude the system prompt, so they may differ slightly from Claude's internal count. The token display becomes invalid after Claude performs a context compaction, which the extension signals visually.

Open source — MIT License.