Claude Usage Monitor

Real-time Claude Code API usage monitor for VS Code — shows cumulative token usage, cache hit rate, and API key balance directly in the status bar.

Features

📊 Cumulative Token Tracking — Automatically scans Claude Code session transcripts (~/.claude/projects/*.jsonl) and aggregates token usage per model
🗄️ Cache Hit Rate — Calculates prompt cache hit rate from cache_read_input_tokens data, with color-coded status bar indicator
💳 API Balance — Queries DeepSeek balance API to show remaining credits
⏱️ Auto-Refresh — Configurable refresh interval (default: 5 minutes)
🔍 Detailed View — Click status bar for per-model breakdown, hover for full stats in markdown tooltip

Backend	Token Tracking	Cache Hit Rate	Balance
DeepSeek (api.deepseek.com)	✅	✅	✅
Anthropic (api.anthropic.com)	✅	✅	❌ (balance requires Anthropic Console API)
Other Anthropic-compatible providers	✅	✅	❌

The extension activates automatically on VS Code startup. Look for the status bar item in the bottom-right corner:

⚡ v4-pro 2.15M tok  🗄 96.2%  💳 ¥23.50

Action	Result
Click status bar	Open detailed QuickPick panel with per-model stats
Hover status bar	Markdown tooltip with full breakdown table
Cmd/Ctrl+Shift+P → `Claude Usage Monitor: Refresh Now`	Manually trigger refresh
Cmd/Ctrl+Shift+P → `Claude Usage Monitor: Show Detailed Usage`	Open detail panel

Indicator	Green	Yellow	Red
Cache hit rate	≥ 80%	50–79%	< 50%

Setting	Type	Default	Description
`claudeUsageMonitor.refreshIntervalMinutes`	number	5	Refresh interval (1–60 min)
`claudeUsageMonitor.showBalance`	boolean	true	Show API balance in status bar
`claudeUsageMonitor.showCacheHitRate`	boolean	true	Show cache hit rate in status bar
`claudeUsageMonitor.claudeDataDir`	string	(auto)	Claude Code data directory path
`claudeUsageMonitor.deepseekApiKey`	string	(auto)	DeepSeek API key for balance check

Token data — Parses every type: "assistant" entry in Claude Code's JSONL session files, extracting input_tokens, output_tokens, cache_read_input_tokens, and cache_creation_input_tokens grouped by model name
Cache hit rate = cache_read_input_tokens / (cache_read_input_tokens + input_tokens)
Balance — Queries GET https://api.deepseek.com/user/balance with your API key
All data stays local — no telemetry, no external servers

This extension: