# How to Track AI Usage Limits and Quotas: TokenBar for Codex, Claude and Cursor > AI Release · @ai_release1 · https://ai-release.org/guides/kak-otslezhivat-limity-i-kvoty-ii-servisov-tokenbar-dly_en.html ## TL;DR - Time to complete: 20–30 minutes - Difficulty: Intermediate - You will need: TokenBar installed from its official documentation, accounts with access to the Codex, Claude, and Cursor usage/billing pages, a terminal with `bash` and `python3`, and a CSV file for your baseline. The source post does not specify TokenBar’s supported OS, exact install command, settings labels, or default refresh interval — check TokenBar’s documentation for those details. ## Requirements and setup ![Illustration: Requirements and setup](https://ai-release.org/guides/img/kak-otslezhivat-limity-i-kvoty-ii-servisov-tokenbar-dly-en-1.jpg?v=1792918856) - OS supported by TokenBar. The source post does not state it; check the official documentation before installing. - Accounts for Codex, Claude, and Cursor that can open each provider’s usage or billing dashboard. Team plans may require admin access; check the documentation. - TokenBar installed from the official source. Record the install path and version in your runbook. - A terminal with `bash`, `python3`, and `date`. On Windows, use WSL or the documented TokenBar alternative. - Write access to `~/ai-quota` for the baseline CSV and verification log. - A shared alert destination, such as a team chat channel or email alias, for quota warnings. TokenBar is the usage tracker named in the source post. It is a monitor, not the source of truth: the provider dashboard remains the authority for quota and billing state. A **quota** is the cap a provider places on usage in a time window; a **baseline** is a dated snapshot you compare later. ## Step 1. Establish a provider-side quota baseline **What to do:** Open the usage or billing dashboard for Codex, Claude, and Cursor. For each tool, record the plan name, quota window, reset time, current usage, dashboard URL, and the UTC timestamp when you read it. Put the values in a CSV. A CSV is a plain-text table with comma-separated columns. ```bash mkdir -p ~/ai-quota cd ~/ai-quota cat > quota-baseline.csv <<'CSV' tool,plan,quota_window,reset_at,used,source_url,checked_at Codex,,,,,, Claude,,,,,, Cursor,,,,,, CSV python3 - <<'PY' import csv with open('quota-baseline.csv') as f: for row in csv.DictReader(f): print(f"{row['tool']:8} used={row['used']:>10} reset={row['reset_at']} checked={row['checked_at']}") PY ``` **Example:** if the Codex dashboard shows a weekly limit and 40% used, write `weekly` in `quota_window` and `40%` in `used`. Do not guess a reset time — copy it from the dashboard. Expected output: ```text Codex used= reset= checked=2025-01-01T00:00:00Z Claude used= reset= checked=2025-01-01T00:00:00Z Cursor used= reset= checked=2025-01-01T00:00:00Z ``` **Why:** TokenBar can show a number, but you need an independent dated record to detect drift, stale data, or a provider-side reset. The CSV also tells you which account and plan you measured. **How to check:** Run `ls -l ~/ai-quota/quota-baseline.csv` and confirm the file exists. Then run the Python block again; you should see exactly three rows: Codex, Claude, and Cursor. ## Step 2. Install and configure TokenBar for Codex, Claude, and Cursor ![Illustration: Step 2. Install and configure TokenBar for Codex, Claude, and Cursor](https://ai-release.org/guides/img/kak-otslezhivat-limity-i-kvoty-ii-servisov-tokenbar-dly-en-2.jpg?v=1792918867) **What to do:** Install TokenBar using the official instructions, then connect the three providers. The source post does not provide an install command, package name, version, or exact menu path, so use the documentation and record what you used. ```bash # Replace this placeholder with the install command from TokenBar's official documentation. ``` Record the install path. If TokenBar installs as a macOS app, an example check is: ```bash ls -ld /Applications/TokenBar.app 2>/dev/null || echo "TokenBar.app not in /Applications; record the actual install path from the documentation." ``` Open TokenBar and go to its settings or connections page. Add Codex, Claude, and Cursor one at a time. If TokenBar asks for credentials, use the provider’s official token or login flow; do not paste account passwords into a third-party field. Set a refresh interval that TokenBar supports without hitting provider rate limits, and set warning and critical thresholds. **Example:** a team policy file with example values — these are your policy, not TokenBar defaults: ```yaml # quota-policy.yaml — team policy, not a TokenBar-native config file providers: codex: warning_at_percent: 70 critical_at_percent: 90 check_every_minutes: 15 claude: warning_at_percent: 70 critical_at_percent: 90 check_every_minutes: 15 cursor: warning_at_percent: 70 critical_at_percent: 90 check_every_minutes: 15 ``` Translate those values into TokenBar’s settings UI. If TokenBar has no threshold setting, keep the YAML as the team policy and check the dashboard manually at the same interval. **Why:** Connecting all three providers in one place gives you a single view of quota pressure. A written policy prevents one team member from setting a 95% critical threshold while another expects an alert at 70%. **How to check:** TokenBar should show one connected row for Codex, one for Claude, and one for Cursor, each with a last-updated timestamp. If a provider row is missing or has no timestamp, the connection is incomplete — re-add it and check the documentation for the required credential scope. ## Step 3. Verify counters, alerts, and fallback ![Illustration: Step 3. Verify counters, alerts, and fallback](https://ai-release.org/guides/img/kak-otslezhivat-limity-i-kvoty-ii-servisov-tokenbar-dly-en-3.jpg?v=1792918882) **What to do:** Compare TokenBar’s displayed value with the provider dashboard after at least one refresh cycle. Log the pair with a timestamp. Then test one alert at the warning threshold, and write the fallback step for stale data. ```bash cd ~/ai-quota cat > quota-check.sh <<'SH' #!/usr/bin/env bash set -euo pipefail tool="${1:?usage: quota-check.sh }" tokenbar_value="${2:?}" provider_value="${3:?}" printf '%s,%s,%s,%s\n' "$tool" "$tokenbar_value" "$provider_value" "$(date -u +%Y-%m-%dT%H:%M:%SZ)" >> "$HOME/ai-quota/quota-check.csv" SH chmod +x quota-check.sh printf 'tool,tokenbar_value,provider_value,checked_at\n' > quota-check.csv ./quota-check.sh Codex "" "" ./quota-check.sh Claude "" "" ./quota-check.sh Cursor "" "" tail -n 4 quota-check.csv ``` **Example:** if TokenBar shows `40%` for Codex and the Codex dashboard shows `40%`, log `Codex,40%,40%,2025-01-01T00:00:00Z`. If TokenBar shows `40%` but the dashboard shows `55%`, the monitor is stale or reading a different quota window. Expected output: ```text tool,tokenbar_value,provider_value,checked_at Codex,,,2025-01-01T00:00:00Z Claude,,,2025-01-01T00:00:00Z Cursor,,,2025-01-01T00:00:00Z ``` Test the warning path: use TokenBar’s test-alert control if it provides one; otherwise temporarily lower the warning threshold in a test profile, wait for one refresh, and confirm the alert arrives in your destination. Restore the real threshold afterward. Fallback rule: if TokenBar’s last update is older than two refresh intervals, use the provider dashboard and log the event in `quota-check.csv`. **Why:** A monitor is useful only when its numbers match the authority and its alerts reach a human. The paired log proves whether a discrepancy is a refresh delay or a configuration error. **How to check:** Run `tail -n 4 ~/ai-quota/quota-check.csv`; you should see three tool rows. Confirm the alert arrived in your channel, then verify that a second alert is not stuck in “test mode.” For teams using Claude Code, pair this quota check with the subagent setup guide: [Claude Code for Teams: Setting Up Subagents and a Shared AI Superagent](https://ai-release.org/guides/claude-code-dlya-komandy-kak-nastroit-subagentov-i-obsc.html?utm_source=tg&utm_medium=channel&utm_campaign=guide_inline&utm_content=guide_to_guide). After quota checks are stable, scan AI-generated code with the [vulnerability guide](https://ai-release.org/guides/kak-proverit-kod-ot-cursor-i-claude-code-na-uyazvimosti.html?utm_source=tg&utm_medium=channel&utm_campaign=guide_inline&utm_content=guide_to_guide). ## Troubleshooting **Problem. TokenBar will not open on macOS** - Error: `“TokenBar” cannot be opened because the developer cannot be verified.` - Cause: macOS Gatekeeper blocked an app downloaded outside the App Store or not yet allowed by the user. - Solution: Open System Settings → Privacy & Security, find the TokenBar block, and choose Open Anyway. If your organization permits the standard quarantine removal command, run it with the real install path: `xattr -dr com.apple.quarantine /Applications/TokenBar.app`, then `open /Applications/TokenBar.app`. If the path differs, use the path you recorded in Step 2. **Problem. A provider connection fails** - Error: `401 Unauthorized` - Cause: the credential is expired, copied with missing characters, or lacks the scope TokenBar needs. A wrong system clock can also break time-based tokens. - Solution: regenerate the credential in the provider’s dashboard, paste it again, and run `date -u` to confirm the system clock is correct. If the error persists, check TokenBar’s documentation for the required credential scope. **Problem. The baseline check cannot find the CSV** - Error: `FileNotFoundError: [Errno 2] No such file or directory: 'quota-baseline.csv'` - Cause: the Python command is running from a different directory, or Step 1 was not completed. - Solution: run `cd ~/ai-quota && ls -l quota-baseline.csv`. If the file is missing, repeat the CSV creation block in Step 1. ## Summary and next steps You now have a dated provider baseline, a configured TokenBar view for Codex, Claude, and Cursor, and a verification log that compares the monitor with the provider dashboards. Next, add the quota check to your team runbook and run the code-security scan linked above on AI-generated changes. ## FAQ **Does TokenBar replace the Codex, Claude, or Cursor billing dashboards?** No. Use TokenBar as a monitor; the provider dashboard remains the source of truth for quota, billing, and reset times. **Which account should I connect for a team plan?** Use an account that can see the team’s usage or billing data. If you cannot see usage in the provider dashboard, ask an admin; TokenBar cannot show data the account cannot access. **How often should TokenBar refresh?** The source post does not specify a default or supported minimum. Choose the shortest interval TokenBar supports without hitting provider rate limits, and verify it against the documentation. **Does TokenBar store prompts or source code?** The source post does not say. Review TokenBar’s privacy and security documentation before connecting a work account, especially if your repositories contain sensitive code.