Claude Code Usage

Claude Code Usage turns your Stream Deck into a live control surface for Claude Code — not just a usage monitor. See your spend, plan limits and session countdown at a glance, then launch a session or watch your background agents without leaving your editor. Since v1.12 it also gets your attention on its own.

ZERO-CONFIG — IT JUST WORKS No API key. No config files. No copy-pasting IDs. The plugin reads your existing Claude Code login and auto-detects your sessions across Windows and every WSL distribution. Drop it on a key and it's live.

SEVEN ACTIONS

• Current Block — Live cost, tokens or percent utilization of your active 5-hour billing window. A progress ring fills as you spend; tap to cycle cost ($1.85) → tokens (380k) → percent (78%).

• Usage Breakdown (new) — One number per key, readable from across the room: your cache hit rate, any of the four token classes, which model is eating your budget (by name, with its cost share), or a burn-rate forecast — how long the budget lasts, or "OK" when your window resets first. Want several at once? Add several keys.

• Usage History — Daily, weekly or monthly bar charts of your spend. Spot the busy days, see trends across months. Tap to switch period.

• Plan Limits — Real-time session and weekly utilization, straight from Anthropic's API. Color-coded green → yellow → orange → red as you approach your cap.

• Session Countdown — Time left in your current 5-hour window with live utilization below it. Keeps counting from local data even if the API is briefly unavailable.

• Claude Launcher — Start or resume a session from a key with your own presets: working directory, permission mode, reasoning effort, initial prompt, extra flags. The key shows a budget traffic light, and Auto mode picks the model for your remaining weekly budget — Opus while you have headroom, then Sonnet, then Haiku. An optional Opus guard asks for a confirm press before an expensive run. Works with Windows Terminal + WSL, plain WSL, PowerShell or cmd.

• Background Agents — A live count of your running background agents: green when they're working, amber when one needs your input. Tap to open the agents view.

STREAM DECK + (new in 1.12) Five actions work as dials, and the touch strip gets its own full-width charts instead of a shrunken key image: spend per hour of your current block, one bar per day with its own figure, both limit bars full width, the token classes as a stacked bar, or every model with its cost share at once.

ALERTS (new in 1.12) Three keys blink to get your attention, then stay lit until you tap them: Background Agents when an agent is blocked waiting for input, Plan Limits when usage crosses a threshold you choose, and Session Countdown when a fresh 5-hour budget is available. Each is switchable in its settings.

WHY YOU'LL LIKE IT ✓ Zero-config — uses your existing Claude Code OAuth token ✓ More than monitoring — launch and steer sessions from your deck ✓ Accurate costs — pricing follows the current Claude model line-up, including both cache-write tiers ✓ Budget-aware — so a long week doesn't silently burn your Opus budget ✓ Privacy-first — session data stays on your machine, and the plugin talks to nothing but Anthropic's own endpoint ✓ Costs in $ or € · settings in English or German · Stream Deck +, MK.2, Mini, XL and Neo

REQUIREMENTS • Stream Deck app 6.9 or later (Windows 10+) • Claude Code installed (any plan)

For live Plan Limits data, run claude setup-token once in your terminal — it makes your existing login long-lived (about a year) and the plugin picks it up automatically.

Claude Code Usage — v1.12.3

Heads-up: your cost figures will change after this update — that's the fix, not a bug. The price table hadn't kept up with the newer Claude models, and unknown ones silently fell back to Sonnet's rates — some were billed far too low, others far too high. Nothing is recalculated retroactively; your existing data is simply priced correctly now. (Sonnet 5 reads ~50% high until 1 Sept 2026, when Anthropic's introductory pricing ends.)

Fixed

Model pricing now resolves by family and version, so future releases land on the right rate automatically. Anything unknown is logged instead of quietly guessed.

1-hour cache writes were charged at the 5-minute rate. They cost twice the input rate, and for heavy use they're the majority.

Both fixes apply to the current block, the history graph and the weekly total.

New

Usage Breakdown key — one number per key: cache hit rate, any token class, which model eats your budget, or how long it will last.

Alerts — Background Agents blinks when an agent needs input, Plan Limits when usage crosses your threshold, Session Countdown when a fresh 5-hour budget starts. Tap to acknowledge; each can be switched off.

Stream Deck+ — five actions work as dials, each drawing a full-width chart on the touch strip instead of a shrunken key image.

Costs in €, configurable traffic-light thresholds, and German settings panels.

Your saved key configurations are untouched