Claude Usage for JetBrains IDEs

A small, free, open-source plugin that shows your Claude subscription usage in the IDE status bar (bottom-right).

The widget shows a Claude icon + 14% used | resets in 1:14 — how much of the selected quota is gone and the time (H:MM) until it resets. Click it for the full breakdown.

Features

• Subscription quota in the status bar — the 5-hour, 7-day and 7-day-Sonnet limits, with a percentage, a colour indicator (green → yellow → red) and time to reset. Same numbers as claude → /usage.
• Tokens used — total tokens sent and received (prompt-cache traffic shown separately) for today / last week / last month, counted locally from the Claude Code transcript logs.
• Click-through popup with the full breakdown and a manual refresh.
• Optional IDE notification when a quota crosses the yellow / red threshold.
• Settings under Settings → Tools → Claude Usage Monitor: which quota to show, refresh interval, colour thresholds, credentials path.
• macOS Keychain support, with a credentials-file fallback.

Install

From the IDE: Settings → Plugins → Marketplace, search "Claude Usage Monitor".

Or grab the ZIP from the Marketplace page / Releases and use Settings → Plugins → ⚙ → Install Plugin from Disk….

Requirements

• An IntelliJ-based IDE 2025.1+ (IDEA, WebStorm, PyCharm, GoLand, Rider, …).
• The Claude Code CLI installed and signed in (claude). The plugin reads the token it stores:
  • macOS: Keychain item Claude Code-credentials, falling back to the file below.
  • Linux / Windows: ~/.claude/.credentials.json.

Privacy

• The credentials are read-only. The plugin never writes them.
• The token is sent to exactly one URL — https://api.anthropic.com/api/oauth/usage (HTTPS, Anthropic's own server), in the Authorization header — to read your usage. Nowhere else. There are no other network calls in the code.
• The token is never logged and never persisted by the plugin; it stays in memory only for the duration of a request.
• The local scanner only reads files under ~/.claude/projects.
• No telemetry, no analytics, no third-party servers.

The relevant code is a handful of files under src/main/kotlin/dev/fvojta/claudeusage/auth/ and api/.

How it works

UsageService (application service, one background thread, refresh every N minutes)
├── CredentialsReader ──► access token (Keychain or ~/.claude/.credentials.json)
├── SubscriptionUsageClient ──► GET /api/oauth/usage ──► 5h / 7d / 7d-Sonnet quota
└── TranscriptScanner ──► walk ~/.claude/projects/**/*.jsonl
        per assistant message: sum usage tokens, bucket by local date,
        cache per file by (size, mtime)
        ▼
   Snapshot ──► ClaudeUsageWidget (status bar) + UsagePopup (details)

Limitations

• /api/oauth/usage is undocumented — it is what claude → /usage calls. Anthropic can change or remove it. The request mirrors the CLI's User-Agent (SubscriptionUsageClient.USER_AGENT) because the endpoint has rejected unknown clients before; bump that constant if the quota stops updating.
• Token counts come from the Claude Code CLI logs only — usage from claude.ai in the browser or other clients is not in those files, so the numbers are a lower bound and won't match the quota percentage exactly.
• Not affiliated with Anthropic.

Development

./gradlew runIde         # sandbox IDE with the plugin loaded
./gradlew buildPlugin     # -> build/distributions/claude-usage-monitor-<version>.zip
./gradlew verifyPlugin    # IntelliJ Plugin Verifier

The build targets a JDK 17 toolchain; the Foojay resolver in settings.gradle.kts provisions it automatically on first run, so no manual JDK install is needed.

License

MIT.