ClaudeBar tracks twenty quota sources, and only three fit in the menu bar

Twenty providers, and each row tracks something different

The provider table is the most informative page in the repository, because the second column does not describe twenty variations on the same number.

Claude tracks a five-hour session window, a weekly limit, per-model limits for three named models, and an extra usage pool. Codex tracks five-hour and weekly rate limits plus credits. Gemini tracks per-model Code Assist quota across three tiers. Copilot tracks monthly AI credits. Cursor splits included monthly usage from on-demand and team credits. Antigravity reports two separate pools with their own five-hour and weekly windows. AWS Bedrock is the odd one out, reporting today's spend and tokens per model with an optional daily budget rather than a subscription allowance at all.

The rest follow the same idea of reporting whatever the vendor actually meters: account balances for DeepSeek, today's spend for Mistral, request counts per model for MiniMax, credit balances for Grok per billing period and per product, and a credit balance for a gateway rather than a coding plan for Vercel.

The consequence is that there is no single percentage the app could compute across all of them. It is a display layer over twenty different vendor APIs, which is why the setup column links to a separate page per provider instead of one shared manual.

The menu bar holds three providers

The menu bar row states the limit plainly: up to three providers shown as a percentage with a reset countdown, or a single status icon. That is the constraint that shapes the whole application.

Twenty integrations compete for three slots, so the configuration question is not which providers you have accounts with but which three you want to glance at. A user who enables everything in the settings panel and then finds only three visible has made the tool less useful, not more, and the page does not say what happens to the remainder.

There is also a status icon mode, which collapses the display to one indicator. That reads as the fallback for a crowded menu bar, and it pairs naturally with the alert levels described elsewhere: healthy, warning, critical and depleted, with colours that take pace into account rather than only the remaining percentage.

The pace-aware part deserves attention. A quota that resets in ten minutes and one that resets in two days can both read as fifty percent and mean very different things, so colouring by pace alone rather than by remaining amount is the more useful of the two designs.

Three features only work on particular MacBook hardware

The features table includes two that depend on what the machine has rather than on the software.

The notch row puts quota and live sessions from Claude Code into the MacBook notch, which is a physical cutout in the display on certain models. The Touch Bar row puts quota gauges on a MacBook Pro Touch Bar, a component Apple has been removing from its line. Both are the kind of feature that makes a screenshot look compelling and then does nothing on the desktop you actually use.

The remaining integrations are platform independent within macOS. Notifications reach an iPhone Lock Screen and Home Screen through a companion app. Session hooks emit started and finished notifications tied to Claude Code sessions, which is the one feature that depends on something outside the app entirely, since a hook has to be installed into another tool for it to fire.

Daily usage is the least exotic and possibly the most useful, showing today's cost and tokens beside yesterday's, which turns the menu from a status display into a running meter.

An extension point reads anything a script can print

The last row of the provider table is not a vendor. It is a directory in the user's home folder, and anything a script can print becomes a quota source.

That is the mechanism that explains how twenty providers were reached without twenty rewrites of the interface. New vendors appear as extensions with a manifest and a script rather than as code merged into the application, and the documentation treats it as a first class path rather than a workaround. The architectural consequence is that the provider list in the table is a snapshot of what shipped, not a hard limit.

The trade is that an extension is a script running with your user's privileges on a machine that also holds your coding tool credentials. The page does not discuss permissions or sandboxing for extensions, which means the trust model is whatever the manifest implies, and a reviewer should read one before installing one.

The same openness appears elsewhere: URL schemes expose actions for launchers and scripts, and themes include importing a terminal's own colour file.

Two install routes, and a source build that needs a toolchain

The quick start is a single command installing a Homebrew cask:

brew install --cask claudebar

The alternative is downloading a signed and notarized disk image from the releases page. The page states the system requirement as macOS 15 or newer and points the setup at a settings pane where each provider is enabled individually, with that provider's own documentation saying what it needs.

That last part is the real setup instruction. There is no single credential to paste, because twenty providers means twenty authentication mechanisms, and the page's decision to delegate each one to its own page keeps the main document short at the cost of spreading the work.

Building from source is delegated to a contributing document rather than described here. The repository layout explains what that involves: a Swift package manifest, a Tuist configuration with its own directory, sources and tests as separate top level directories, a modules directory, and a scripts directory. Tuist is a build system for Xcode projects, so the project is generated rather than opened as a checked in workspace, which is the pattern the directory listing implies.

Releases move fast by conventional standards, with three tags in the last week at the time of writing, and the newest is on the release page the install link points to.

Apache 2.0 and third-party terms

The repository includes a LICENSE file that GitHub identifies as Apache-2.0. The project also links to contribution, changelog and architecture documentation, publishes signed and notarized builds, and provides a package-manager cask.

If you distribute or modify the application, review the LICENSE file and any applicable third-party terms. Extension scripts can be third-party code and may carry separate licenses.

Multiple accounts are handled for one provider only

The features table has a row for multiple Codex accounts, and it is the only entry of its kind. Separate logins are identified by email, each with independent quotas.

That asymmetry is a reasonable response to how people actually work. It is common to hold two coding plan accounts, for example one for personal work and one billed elsewhere, and the feature keeps them apart rather than summing them into a single misleading number. The cost is that the same treatment is not offered for the other nineteen providers, where multiple accounts would presumably be aggregated or ignored.

The identity mechanism is worth noting too. Accounts are keyed by email rather than by a token or a workspace identifier, which is simple and readable, and means switching identity depends on which credentials are present rather than on a selector in the interface.

For the providers where a balance rather than a window is reported, the same question applies differently, since two accounts with balances would produce two figures and there is no stated rule for combining them.

Editorial conclusion

ClaudeBar is worth installing if you juggle more than one AI coding subscription and want to know which one is nearly spent before a long task starts. Its real strength is the per-provider documentation, since each of the twenty integrations states what credentials it needs, and that is the part you will actually consult. Two cautions. The menu holds three providers, so turning on everything means the ones you care about will not all be visible at once, and you should pick your three before you enable anything. And the app is macOS 15 or newer with several features that only work on particular MacBook hardware, so check what your machine supports before planning a workflow around the notch or the Touch Bar.

Frequently asked questions

Official sources

1. Issues
2. Project website
3. README
4. Releases
5. tddworks/ClaudeBar on GitHub