Self-hosted dashboard to monitor your Claude.ai usage limits in real time. Polls the claude.ai internal API every 60 seconds and displays your 5-hour session and 7-day rolling utilization with a history graph — accessible from any device on your home network.
- Live usage bars — 5-hour session and 7-day rolling utilization
- Color-coded alerts — green → yellow → red as you approach limits
- Reset countdowns — exact time until each window resets
- History graph — view usage over 3h, 6h, 24h, or 7 days
- Adjustable poll interval — change how often the server checks Claude.ai, live from the Settings panel (1–15 min), no restart needed
- Rate-limit safe — 60-second minimum floor, randomized jitter, and automatic backoff if Claude.ai returns an auth or rate-limit error
- SQLite persistence — history and your interval setting survive container restarts
- Lightweight — ~150 MB Docker image, no browser automation needed
- Accessible — WCAG 2.2 AAA oriented, full keyboard navigation, screen-reader summaries
- LAN accessible — open on any phone, tablet, or laptop on your network
Claude.ai is a React app that fetches usage data from an internal API endpoint. ClaudeDash calls that same endpoint using your session cookies, caches the result, and serves it as a clean dashboard.
Every N minutes (default 3, configurable):
GET https://claude.ai/api/organizations/{org_id}/usage
↓
{ five_hour: { utilization: 41.0, resets_at: "..." },
seven_day: { utilization: 12.0, resets_at: "..." } }
↓
Saved to SQLite → served at http://your-server:3003
No Playwright, no headless browser, no Anthropic API key needed.
ClaudeDash is deliberately conservative about how often it calls Claude.ai:
- Default interval is 3 minutes. Usage numbers don't move fast, so frequent polling buys nothing.
- Hard floor of 60 seconds. The server clamps any requested interval so it can never poll faster than once a minute, even if you edit the value directly.
- Jitter. Each poll fires at the interval ± a few seconds of random offset, so requests don't arrive in perfect clockwork (which looks more automated).
- Automatic backoff. If Claude.ai responds with 401, 403, or 429, the server doubles its interval (up to a 30-minute cap) instead of retrying aggressively, then automatically returns to your configured interval on the next successful poll. The dashboard shows a "Backed off" badge while this is active.
You can change the interval any time from Settings → Poll Interval; the choice is saved to SQLite and applied immediately.
- Docker + Docker Compose
- A Claude.ai account (Pro or Max)
- Google Chrome (to extract session cookies once)
git clone https://github.com/yourusername/ClaudeDash.git
cd ClaudeDashOpen claude.ai in Chrome, then press F12 → Application tab → Cookies → click https://claude.ai.
Find and copy these values:
| Cookie Name | .env Variable |
|---|---|
sessionKey |
CLAUDE_SESSION_KEY |
lastActiveOrg |
CLAUDE_ORG_ID |
anthropic-device-id |
CLAUDE_DEVICE_ID |
ajs_anonymous_id |
CLAUDE_ANON_ID |
cf_clearance |
CLAUDE_CF_CLEARANCE (optional) |
__cf_bm |
CLAUDE_CF_BM (optional) |
Note: For
ajs_anonymous_id, there may be two entries — pick the one with domain.claude.ai(value starts withclaudeai.v1.).
cp .env.example .envEdit .env and paste your cookie values:
CLAUDE_SESSION_KEY=sk-ant-sid02-...
CLAUDE_ORG_ID=xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx
CLAUDE_DEVICE_ID=xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx
CLAUDE_ANON_ID=claudeai.v1.xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx
CLAUDE_CF_CLEARANCE=optional
CLAUDE_CF_BM=optionaldocker compose up --build -dOpen http://localhost:3003 — or http://your-minipc-ip:3003 from any device on your LAN.
ClaudeDash/
├── Dockerfile
├── docker-compose.yml
├── main.py # FastAPI backend + APScheduler poller
├── requirements.txt
├── public/
│ └── index.html # Dashboard UI (Chart.js, vanilla JS)
├── data/ # SQLite database (created at runtime, gitignored)
├── .env.example
├── .env # Your cookies — never commit this
└── .gitignore
| Variable | Default | Description |
|---|---|---|
POLL_INTERVAL |
180 |
Initial poll interval in seconds, used only the first time the database is created. After that, change it from the Settings panel. Clamped to a 60s minimum. |
DB_PATH |
/app/data/usage.db |
SQLite database path (mounted volume for persistence) |
The poll interval is stored in SQLite once set, so editing POLL_INTERVAL in docker-compose.yml after first run has no effect — use the UI instead.
| Method | Path | Purpose |
|---|---|---|
GET |
/api/usage |
Latest cached usage reading |
GET |
/api/history?hours=N |
Usage history for the last N hours |
GET |
/api/settings |
Current interval, effective interval, backoff state |
POST |
/api/settings |
Set the poll interval ({ "interval": 180 }) |
POST |
/api/poll-now |
Trigger an immediate poll |
Session cookies expire every few weeks. When the dashboard shows an auth error (or the "Backed off" badge appears and stays):
- Re-open Chrome → F12 → Application → Cookies → claude.ai
- Copy the new
sessionKeyandcf_clearancevalues - Update your
.env - Restart:
docker compose restart
| Layer | Technology |
|---|---|
| Backend | Python 3.12, FastAPI, APScheduler |
| HTTP client | httpx |
| Database | SQLite |
| Frontend | Vanilla JS, Chart.js |
| Fonts | Bebas Neue (numbers), Syne (headings), DM Mono (UI) |
| Container | Docker, ~150 MB image |
- Never commit
.env— yoursessionKeyis equivalent to your Claude.ai login - The app runs entirely on your local network — do not expose port 3003 to the internet
- Cookies are read-only; ClaudeDash never writes to your Claude account
- Breaks if Anthropic changes the internal API endpoint or response format
- Session cookies expire and need manual renewal every few weeks
__cf_bm(Cloudflare) cookie rotates every ~30 minutes — safe to omit
- Clawdmeter by @HermannBjorgvin — ESP32 physical desk dashboard
- ClaudeMeter by @eddmann
MIT