Claude Code Usage Meter

A VS Code extension that puts your Claude Code usage in the status bar: subscription session and weekly limits, context window fill, cost, and burn rate — updating live via file watcher, no polling.

What's new in 2.0.0

Claude Context Meter is now Claude Code Usage Meter, with a new logo. It started as a context-window meter; it now tracks subscription limits, cost, burn rate and spend too, so the name follows what it actually does. Same extension — it updates in place, and your settings and keybindings are untouched.

Subscription usage (added in 1.6.0) shows your Claude session (5-hour) and weekly limits in the status bar, matching what /usage reports in Claude Code:

⏱ Usage 45% · 6%w        🤖 my-project ███░░ 16% ~$1.20 🔥2.1k/m

Crossing a threshold raises a one-time notification, so a cap that stops your work doesn't arrive unannounced. Full history in the changelog.

Status bar

🤖 my-project ███░░ 156k/200k (78%) ~$1.20 🔥2.1k/m

The bar fills as your context fills. After the percentage it shows the running session cost (~$1.20) and the recent burn rate (🔥2.1k/m = ~2,100 tokens/min). Color changes automatically:

• Default — under warning threshold
• Yellow — above warning (default 50%)
• Red — above danger threshold (default 75%)

You also get a warning/critical notification the first time a session crosses each threshold.

Click actions

Click a meter to open a quick menu for that session:

• Hide — dismiss this meter until the session has new activity
• Open transcript — open the session's .jsonl log
• Copy stats — copy this session's stats as a markdown row

Tooltip (hover)

my-project

`claude-sonnet-4-6`  ·  🟡 warn · 78%

████████████████░░░░  78%
156,121 / 200,000 tokens

───────────────────────────────────
in: 140,000  ·  out: 16,000  ·  cr: 0  ·  cw: 121

🔥 recent: ~2.1k/min  ·  avg: ~2.1k/min
⏳ ~21 min to full

💰 cost: $1.20  (in: $0.42 · out: $0.24)

Updated 9:42:15 PM

Subscription usage

⏱ Usage 45% · 6%w

Shows how much of your Claude plan you've used: the session (5-hour) window and the weekly limit, matching what /usage reports in Claude Code. Bars run blue while you have room, amber past usageWarningThreshold, and red past usageDangerThreshold; the status bar item picks up a matching warning or error background. Hovering shows every limit your plan reports:

Claude subscription usage

Session  ·  🟢 safe

🟦🟦🟦🟦🟦⬜⬜⬜⬜⬜  45% used
Resets 7:39pm

Week (all models)  ·  🟢 safe

🟦⬜⬜⬜⬜⬜⬜⬜⬜⬜  6% used
Resets Aug 24, 3:59pm

───────────────────────────────────
Updated 2 min ago

Crossing a threshold raises a one-time notification per limit, so a cap that stops your work doesn't arrive unannounced. Each limit alerts at most once per window and re-arms when that window resets. Alerts only fire on live readings — a stale cached figure never triggers one.

How it works

Usage is read from GET /api/oauth/usage — the same request Claude Code makes for its own /usage command — authenticated with the sign-in token Claude Code already stores on this machine (~/.claude/.credentials.json, or the OS credential store). The token is used only as that request's Authorization header, is sent only to Anthropic, and is never stored or logged by the extension. Refreshes every usageRefreshInterval seconds (default 60).

Claude Code also caches its last usage response in ~/.claude.json, and that cache is used as a fallback whenever the request fails — offline, rate-limited, expired token, or the endpoint changing. The cache refreshes rarely: it has been observed sitting over 30 minutes stale, reporting 45% while actual usage was 85%. So cached readings are labelled as cached, and flagged with ⚠ once older than usageStaleMinutes, rather than being presented as current.

To avoid network requests entirely, set claudeContextMeter.usageLiveFetch to false. Usage then comes from the local cache alone — accepting that it will often be out of date.

Note: this endpoint is undocumented. It may change or stop working at Anthropic's discretion, in which case the meter falls back to the cache rather than erroring.

If you use an API key rather than a Claude subscription there is no usage data, and the item simply doesn't appear. Set claudeContextMeter.showUsage to false to hide it.

Commands

Available from the Command Palette (Ctrl/Cmd+Shift+P):

• Claude Code Usage Meter: Copy Context Stats — copy a markdown table of all active sessions to the clipboard
• Claude Code Usage Meter: Show Spend Summary — pick Today / This Week / All Time and see total cost across every session (active or idle), with a per-project breakdown you can copy as markdown
• Claude Code Usage Meter: Show Session Detail — same quick menu you get by clicking a status bar item (Hide / Open transcript / Copy stats)
• Claude Code Usage Meter: Show Subscription Usage — list every plan limit with its reset time
• Claude Code Usage Meter: Show Diagnostics — running version, detected model, resolved pricing tier and rates, context limit, and transcript paths. Handy when reporting an issue

Budget alerts

Set claudeContextMeter.dailyBudget to a USD amount and you'll get a one-time warning notification the first time that day's total spend (across all sessions) crosses it. Re-arms automatically at midnight. Default 0 disables it.

How it works

Claude Code writes session logs as .jsonl files under ~/.claude/projects/. This extension watches that directory with VS Code's file system watcher — no polling interval, no delay. The moment Claude writes a response, the meter updates.

Token limit is auto-detected from the model name:

New Claude releases default to 1M unless they are Haiku — no extension update required. Cost is priced per model tier and kept current as Anthropic ships new pricing.

Also works with Bedrock, Vertex AI, and Foundry-style model IDs (e.g. anthropic.claude-opus-5-v1:0, claude-opus-4-5@20251101) — pricing, context detection, and the tooltip's cleaned-up model name all handle the provider prefixes/suffixes.

Unrecognized non-Claude models fall back to the contextLimit setting.

Install

From the Marketplace (recommended)

Install Claude Code Usage Meter from the VS Code Marketplace, or search "Claude Code Usage Meter" in the Extensions panel (Ctrl/Cmd+Shift+X). Updates arrive automatically.

From VSIX

Use this if you're on VSCodium, Cursor, or another build without Marketplace access — or if you're testing a pre-release build.

1. Download the latest .vsix from Releases
2. In VS Code: Extensions → ··· → Install from VSIX
3. Click "Reload Now" when prompted, or run Developer: Reload Window. Without a reload the old version keeps running and the install looks like it did nothing.

Or via terminal:

code --install-extension claude-context-meter-2.0.1.vsix --force

Heads up: VS Code pins extensions installed from a .vsix, which turns off automatic updates for them. You'll stay on that exact version until you update by hand. To go back to automatic updates, install the Marketplace copy instead — note this is the extension ID, not a file path:

code --install-extension kexgev.claude-context-meter

Run Claude Code Usage Meter: Show Diagnostics from the Command Palette at any time to confirm which version is actually active.

Build from source

git clone https://github.com/kexgev/claude-context-meter.git
cd claude-context-meter
npm install
npx vsce package --allow-missing-repository
code --install-extension claude-context-meter-2.0.1.vsix --force

On Windows, ./scripts/install-local.ps1 does the build, install, registration check, and stale-folder cleanup in one step.

Requirements

• VS Code 1.74+
• Claude Code CLI

Settings

License

MIT