Claude Monitor

Real-time dashboard for tracking Claude Code usage, costs, and session activity. Built with Bun, React, and SQLite.

Features

• Overview Dashboard — CodeBurn-inspired dense, multi-panel layout showing cost, tokens, sessions, cache hit, daily activity, projects, models, activity types, tools, and MCP servers — all visible at once
• Live Sessions — Real-time view of all running Claude Code instances with CLI/Desktop icons
• Rate Limits — Session and weekly usage percentage tracking with color-coded bars (60s auto-refresh)
• Session History — Per-session metadata table with model, effort level, tokens, cost; click any row for full tool call + prompt timeline
• Usage Statistics — Today/7 Days/30 Days breakdowns with period comparison badges
• Cost per Project — Projects ranked by cost with automatic subdirectory merging
• Model Breakdown — Token distribution across Opus, Sonnet, and Haiku
• Activity Categories — Tool calls grouped into Coding, Exploration, Testing, Planning, Git, and Delegation
• Tool Usage — Per-tool call counts, durations, error rates, and over-time charts powered by OTEL tracing
• MCP Server Analytics — Dedicated panel for MCP tool usage tracking
• Thinking Depth — Tracks extended thinking frequency and depth across sessions
• macOS Menu Bar App — Native Swift app (Claude Code Monitor.app) with live stats, server controls, and quick dashboard access
• OTEL Collector — Built-in OTLP/HTTP endpoint on ports 4500 and 4318 for ingesting Claude Code telemetry

Quick Start

# Install dependencies
bun install

# Start the dashboard (port 4500)
PORT=4500 bun run dev

# Or use the start script (background process)
PORT=4500 bun run start

Open http://localhost:4500

How It Works

The monitor watches ~/.claude/projects/ for JSONL session logs written by Claude Code. Each API response with token usage is parsed, priced, and stored in SQLite. A WebSocket pushes live updates to the dashboard.

Data Sources

Source	What it provides
~/.claude/projects/**/*.jsonl	Per-request token usage, model, session ID
~/.claude/sessions/*.json	Live session PIDs and entrypoints (CLI vs Desktop)
~/.claude/usage-data/session-meta/	Historical session metadata
~/.claude/stats-cache.json	Daily activity and model token stats
Status line hook	Rate limit percentages (session/weekly)
OTEL collector (port 4318)	Metrics, events, and distributed traces

Scripts

Command	Description
bun run dev	Start with hot reload
bun run start	Start as background process
bun run stop	Stop background process
bun run build	Compile to standalone binary
bun run backfill	Import historical data from JSONL logs

macOS Menu Bar App

Install the native menu bar app for quick access:

cd ClaudeMonitorBar
./build.sh
cp -r "build/Claude Code Monitor.app" /Applications/
open "/Applications/Claude Code Monitor.app"

The app shows live session count and token burn rate in the menu bar, with controls to start/stop/restart the server and open the dashboard.

Enable Tool Usage Analytics (OTEL)

The monitor listens on port 4318 (standard OTLP default) — no endpoint config needed. Just enable tracing in your shell:

# Add to ~/.zshrc (the monitor writes this file on every start)
source ~/.claude-monitor.env

Or set manually:

export OTEL_LOG_TOOL_DETAILS=true       # required: per-tool call stats
export OTEL_LOG_TOOL_CONTENT=true       # optional: capture input/output
export OTEL_LOG_USER_PROMPTS=true       # optional: capture prompt text

Restart your terminal and run claude — the Tool Usage tab will populate with real tool call data.

Tech Stack

• Runtime: Bun
• Frontend: React 19 + Recharts
• Database: SQLite (WAL mode)
• Telemetry: OpenTelemetry OTLP/HTTP

License

MIT