See your Claude Code limits without typing /usage
A small always-on-top Windows widget that keeps your Claude usage on screen: the 5-hour session limit, the weekly limit, how much you can still spend today, and how many tokens you have burned.
Free and open source (MIT) · Windows 10 and 11 · works with Claude Pro and Max · not affiliated with Anthropic
Example numbers. Click a theme to try it.
A Claude Code usage monitor that stays on screen
Claude Code tells you your usage when you type /usage. That interrupts the agent, and you only find out you hit the rate limit when it stops working. The widget shows the same numbers all the time, and adds the ones /usage does not have.
Session and weekly limit bars
The 5-hour session limit, the weekly limit and any per-model weekly limit, each with the time until it resets. Amber at 60 %, red at 85 %.
Daily budget
How much of the weekly limit you can spend today so that you neither run out early nor leave it unused.
Pace forecast
"On pace for 96 % by the reset" for the week, and the time your current session will be full if you keep going at this speed.
Token usage today
Input, output and cache tokens, the number of replies and the burn rate over the last hour, read from your local Claude Code transcripts.
What it would cost on the API
Your usage priced at API list rates, per day, per model and per session. Shows what the subscription is worth to you.
Models, subagents, 7 days
The split by model, the share burned by subagents, the most expensive sessions of the day and a chart of the last week.
F9 terminal hotkey
One key brings the Claude Code terminal to the front or puts it away. Shift+F9 opens another session as a tab.
Yours to arrange
Switch every block on or off. Four themes, adjustable transparency, any size. Drag it anywhere; it remembers.
Use the whole weekly limit, without running out
A weekly limit that resets with 40 % unused is wasted. One that is gone by day five stops your work. The daily budget splits what is left evenly over the days until the reset.
- The dashed slice is today's share of the week.
- The white tick is where the bar should be by midnight.
- Unused budget rolls over into the following days automatically.
- The line turns amber when you spend past today's share.
F9: the Claude terminal as a drop-down
With many windows open, the Claude Code terminal is the one you keep hunting for. The widget registers a global hotkey for it.
The key can be changed at the top of the script.
Private by design
- Reads the login Claude Code already has and never refreshes or stores it.
- One request, to api.anthropic.com, the same one
/usagemakes. - Token statistics stay local: they are counted from the transcripts on your PC.
- Two readable source files, no installer, no background service, no telemetry.
Install in one minute
You need Windows 10 or 11, Claude Code logged in with a Claude subscription, and Windows Terminal for the F9 hotkey. PowerShell 5.1 is already part of Windows.
git clone https://github.com/F3S0J/claude-usage-widget.git cd claude-usage-widget powershell -ExecutionPolicy Bypass -File .\install-shortcuts.ps1
This puts a Claude Usage shortcut on the Desktop and in the Startup folder, so the widget starts with Windows. No git? Download the ZIP, unpack it and run the same last line in that folder.
Questions
How do I check my Claude Code usage limit without typing /usage?
Run the widget. Every two minutes it asks the same server endpoint that /usage uses, and keeps the 5-hour session limit and the weekly limits on screen as bars with their reset times.
Which Claude plans does it work with?
It needs Claude Code logged in with a Claude subscription (Pro or Max). It shows whichever limits your plan returns, including per-model weekly limits.
Is my Claude login token safe?
The widget reads the token Claude Code already stores on your PC and sends it only to api.anthropic.com, in the same request Claude Code makes. It never refreshes or stores the token. The code is one PowerShell file and one C# file you can read.
How is the dollar value calculated?
It multiplies the tokens found in your local Claude Code transcripts by the API list prices per model, including cache reads and cache writes. It is a comparison with pay-as-you-go API pricing, not a bill. The prices are written in the code and need updating when they change.
Do the token statistics include claude.ai chats?
No. The limit bars come from the server and cover everything on your account. The token counts, cost and charts come from the Claude Code transcripts on this PC only.
Does it work on macOS or Linux?
No. It is built on Windows PowerShell and WPF and runs on Windows 10 and 11 only.
Why does the widget say "limited" or "offline"?
"Limited" means the usage server answered 429 Too Many Requests; the widget shows the last saved numbers and retries later. "Offline" means a real error, most often an expired token, which fixes itself the next time you open Claude Code.
Is this an official Anthropic tool?
No. It is an independent open-source project. The usage endpoint it reads is the one the official client uses, but it is not a documented public API and can change without notice.