Claude Code Usage

The Claude Code coach in your status bar. Not a billing tool. Not a multi-provider monitor. A focused token tracker that uses AI to help you use Claude Code better.

What this is: A VS Code status-bar monitor that reads your local Claude Code conversation logs and shows token-derived usage and cost estimates — plus an optional AI advisor that suggests how to improve your prompts and reduce waste.

What this is not: a billing tool. All amounts are estimates based on public per-million-token rates. Refer to your Anthropic account for actual charges.

看清你的 Claude Code 用量，让 AI 帮你用得更好。

简介：一个 VS Code 状态栏小工具，读取本地 Claude Code 对话日志， 按 token × 公开单价估算用量与成本；并提供可选的 AI 建议功能，帮你优化 提示词、减少不必要的 token 消耗。

它不是什么：账单工具。显示金额均为估算值，实际费用请以官方账单为准。

🌐 Multi-language documentation: English · 繁體中文 · 简体中文 · 日本語 · 한국어

Today's cost · current-session cost · 5-hour and weekly quota utilisation.

Hover the quota indicator for a breakdown:

Real /usage data — utilisation percent, reset countdown, and the weekly reset weekday and time.

Click the status bar to open the full dashboard. Stacked token-composition chart, hourly breakdown, cache hit rate, cost composition by token type, plus per-model and per-day tables below.

Estimated breakdown of which content consumes tokens — your prompts vs. tool results (by tool) vs. assistant output / thinking. This is the lever for optimising your usage. Scoped to the last 30 days (advice.promptWindowDays).

AI advice writes you a Markdown document, so it reads better as text than as a screenshot. Set a key (advice.apiKey), click Get AI advice (the ✨ button or the card on the Content tab), pick a scope (all projects, or one), and it sends your usage aggregates + a sample of your own prompts to your model and opens a prioritised report. Bring your own key — Anthropic (/v1/messages) by default, or any OpenAI-compatible endpoint.

A flavour of what it returns (illustrative):

Write more complete instructions

• Several prompts open with "fix the bug" but don't name the file or the symptom, so the first turn is spent searching. Lead with the file + expected vs. actual behaviour.

Cut waste where it doesn't cost clarity

• ~38% of your tokens are spent above 150k context. /clear between unrelated tasks keeps each request cheaper.

Paste a rough, half-formed request; get back one clean, paste-ready prompt (plain text, no Markdown) plus a recommended reasoning effort / thinking / model shown as chips. Three optional toggles refine it (flag vague references · condense long pastes · suggest a style direction). Experimental, off by default; only the text you paste is sent — never your files or the terminal — behind a one-time consent prompt.

• Sessions: resume / copy / delete — each row can copy the session id, resume it (official Claude Code extension in-tab for this project, or a terminal claude --resume <id> for other projects), or delete it (to the trash, with confirm). A Current project / All filter defaults to the current project.
• Quota display options — quotaFiveHourOnly (show only the 5h window) and showResetInStatusBar (append the reset countdown, e.g. 5h:50%:2.3h | wk:30%:3.2d), both in ⚙ Settings.
• Wider dashboard — detail page widened to 1600 px, still fluid on narrow screens.
• Workflows tab — every multi-agent run in one place: dynamic-workflow runs (ultracode) and ad-hoc sub-agent batches, with per-run cost, agent count, models used, cache hit rate (the "is my provider workflow-ready" diagnostic) and a per-agent drill-down labelled by each agent's task.
• Usage tracking panel — the official /usage "what's contributing" view, but multi-provider and with five scopes (Day / Week / Month / session / project): >150k-context share, 8h+-session share, subagent-heavy share, workflow share, plus Skills / Subagents / Plugins / Models breakdowns. Compact card on the Today tab.
• Thinking share per session (Sessions column + Today card) with an /effort hint when it runs high.
• Workflow quota guard — a dismissible banner before you start a run the remaining 5-hour window can't finish (claudeCodeUsage.workflowQuotaWarnPercent).
• Settings in the dashboard — a new ⚙ Settings tab manages every option in place; VS Code's own Settings keeps only the three that benefit from syncing (language, dataDirectory, advice.apiKey). Header buttons trimmed to ✨ AI advice and ⚙ Settings (both jump to their tab); the auto-refresh toggle moved into Settings (a manual ↻ appears when paused). If you hide the cost, quota and context items, the status bar keeps a small icon as a way back into the dashboard.
• Status-bar metric (statusBarMetric) — keep showing today's cost, or switch the first item to today's total token count (compact k/M).
• Weekly Opus limit (showOpusWeekly, opt-in) — append opus:NN% to the quota item for heavy Opus users. (PR #38, @wheelbarrel00.)
• AI advice 2.0 — bring your own key: Anthropic (/v1/messages) by default, or any OpenAI-compatible endpoint (advice.apiFormat). Fed with the new signals (runs, cache hit rates, attribution, thinking share); optional advice.userContext adds a "Personalised for this project" section; advice.promptWindowDays (default 30) sets the sampling window. Transport hardened: timeout, retry, curl fallback. (A keyless "subscription" backend was prototyped but isn't shipped — Anthropic blocks calling the API with the Claude Code OAuth token; it may return if that changes.)
• Usage Optimizer (experimental, advice.optimizer.enabled, default off) — a Content-tab card where you paste a rough request and get back one tightened prompt as plain text (paste-ready, no Markdown) plus a recommended effort / thinking / model. Three optional lenses (flag ambiguous references · condense long pastes · suggest a style direction). Only the text you paste is sent, behind a one-time consent prompt.
• Context-window indicator (experimental, off by default) — opt in via Settings to show the current session's context fill in the status bar. A "~" marks a guessed window; set contextWindowOverride for proxied/custom models.

• Real 5-hour and weekly quota in the status bar — reads Claude Code's existing OAuth session from ~/.claude/.credentials.json or the macOS Keychain, zero config. Adapted from upstream PR #9 by @Dobidop.
• Four new tabs: Sessions, Projects, Content, Branches — all sortable.
• Token-composition stacked chart with Y-axis and reference lines.
• AI advice command (DeepSeek V4 Pro default, reasoning_effort=max) with a demo-mode fallback when no API key is configured.
• Multi-vendor pricing: Opus 4.x, Sonnet 4.x, Haiku 4.5 (verified against Anthropic's public pricing); reference rates for proxied setups (OpenAI, Gemini, DeepSeek, Kimi, GLM, Qwen) with family-aware fallback. Refresh Token Pricing pulls live LiteLLM data as runtime overrides.
• Custom timezone for date display (claudeCodeUsage.timezone).
• Light-theme tab readability fixed.
• Locale-aware numbers and dates throughout (German ., English ,).
• Real-time status bar via fs.watch (1.5 s debounce) + idle-aware refresh + non-blocking loader (yields every 25 files).

Full changelog: CHANGELOG.md. Closes upstream issues #7, #10, #11, #13.

Search for Claude Code Usage in the Extensions view (Ctrl+Shift+X), or:

ext install GrowthJack.claude-code-usage

Same extension is published at the Open VSX Registry: GrowthJack.claude-code-usage.

Ctrl+Shift+P → Extensions: Install from VSIX... → pick the downloaded .vsix.

Most settings live in the dashboard now. Open the dashboard (run Show Usage Details, or click the ⚙ in its header) and use the ⚙ Settings tab — grouped into General, Status bar, Data & refresh, and AI advice & Optimizer. Changes apply immediately.

To keep VS Code's own Settings UI uncluttered, only three settings stay there (so they still travel with Settings Sync). Open Settings (Ctrl+,) and search for Claude Code Usage:

Everything else — refresh interval, status-bar items, number/date formatting, project grouping, content analysis, and all the AI advice / Optimizer options — is in the dashboard's ⚙ Settings tab. Upgrading keeps your existing values: a one-time migration copies them out of settings.json on first launch.

The status-bar cost is Σ (tokens × per-million rate) across input, output, cache-write and cache-read, summed by model.

• Per-million rates come from the bundled pricing table, which is verified against the public Anthropic pricing page and supplemented with reference rates for non-Anthropic models that may appear in proxied setups.
• Refresh Model Pricing (command + button in the dashboard) pulls live prices from LiteLLM's public dataset as runtime overrides.
• Unknown model snapshots are priced against the current tier of their detected family (Opus / Sonnet / Haiku / GPT / Gemini / DeepSeek / Kimi / GLM / Qwen) instead of falling back blindly.

What the status bar does not know:

• Your actual Anthropic invoice (discounts, free credits, plan caps).
• Whether your proxy provider charges different rates.
• Anything not recorded in your local .jsonl log files.

The 5h / weekly quota indicator is different — it queries Claude Code's real /usage endpoint via the OAuth session and shows the actual percentage Anthropic is tracking for your account. That number is authoritative.

• All token / cost / session analysis runs locally by reading your ~/.claude/projects/**/*.jsonl files.
• The quota indicator calls api.anthropic.com/api/oauth/usage using Claude Code's existing OAuth token. No additional credentials are sent.
• AI advice and the Usage Optimizer are the only features that call a model — and only when you trigger them. AI advice sends an aggregate summary of your usage plus a sample of your recent prompts; the Optimizer sends only the text you paste into it (never your files or the terminal), behind a one-time consent prompt. Both send to the endpoint in advice.apiUrl with your own advice.apiKey (Anthropic /v1/messages by default, or any OpenAI-compatible endpoint). Bring your own key; nothing is shipped with the extension.

"No Claude Code Data"

• Make sure Claude Code is installed and you have used it at least once.
• Check the dataDirectory setting; auto-detection looks at ~/.claude/projects and ~/.config/claude/projects.

Quota row shows 5h:--% wk:--%

• Claude Code's OAuth token is missing or expired. Log in to Claude Code once; the extension reads ~/.claude/.credentials.json where present, or the macOS Keychain entry used by Claude Code, and refreshes the bearer if needed.

Get AI Usage Advice returns 404

• DeepSeek's current endpoint does not use a /v1 prefix. Use https://api.deepseek.com/chat/completions. The extension auto-strips /v1 if present.

Get AI Usage Advice shows demo instead of real advice

• AI advice needs a key. With no key under claudeCodeUsage.advice.apiKey, the command opens a hand-written demo (filename-marked …-DEMO-…, with a prominent banner) instead of calling any API. Add a key in Settings to get real advice.

Sluggish refresh on large histories

• 2.0 yields to the event loop every 25 files; idle ticks skip recompute. If you still hit issues, raise refreshInterval or set enableContentAnalysis to false.

Usage history disappears or is missing older months

• Claude Code automatically deletes conversation logs older than cleanupPeriodDays (default: 30 days). Once deleted, those records cannot be recovered. To retain more history, add this to your ~/.claude/settings.json:

  { "cleanupPeriodDays": 365 }

  This only affects logs kept from now on; already-deleted logs cannot be restored. Thanks to @nickearnshaw for documenting this (PR #21).

Token counts appear lower than the model provider's own dashboard

• If you use Claude Code with a third-party proxy that routes requests through sub-agents or background workflows (e.g. ultracode / dynamic workflows), each agent writes its own .jsonl log file inside a sub-directory. The extension reads all these files, but some proxy configurations may not write agent-level records at all. Until native workflow attribution is added in a future release, the total shown here may be lower than the provider's upstream count. Your actual spend is always on your provider's billing page.

Maintained by @Carl723000, who forked it from @jack21's original ClaudeCodeUsage and now also helps own and maintain the upstream organization ClaudeCodeUsage/ClaudeCodeUsage. MIT-licensed. The 2.x work documented here (everything under "What's new") is by @Carl723000 with Claude Code; it has grown well beyond the 2.0 baseline — see CHANGELOG.md.

Contributors whose upstream PRs / issues are incorporated here:

• @Dobidop — PR #9, the OAuth approach for reading real /usage data; the quota indicator is adapted from that work.
• @nickearnshaw — PR #8 locale-aware number/date formatting; PR #20 fix for the webview/status-bar getting stuck on "Loading…" (re-entrancy guard + spinner only on cold start); PR #21 docs on cleanupPeriodDays for retaining usage history; PR #24 quota-window rollover handling (drop a window once its reset has passed).
• @ScherbakovAl — PR #31, the original status-bar context-window indicator and the showCost toggle.
• @wheelbarrel00 — PR #38, the opt-in weekly Opus limit (showOpusWeekly) in the status bar.
• @brenoneill — PR #14, custom data directory (merged into upstream 1.0.8).
• @mxzinke — Opus 4.5 / Haiku 4.5 prices
  • German translation (upstream 1.0.8).

Also closed along the way: the test-suite seed (#25) and unreliable context-window detection for proxied/custom models (#31).

Many code changes in this fork were drafted with assistance from Claude Code (commits include Co-Authored-By: Claude <noreply@anthropic.com>).

The current changelog lives in CHANGELOG.md. The most recent 2.1 entry summarises every feature, fix and personalisation option in this release.

Issues and pull requests are welcome on the GitHub repository.

MIT