Claude usage tracker browser extensions: what they can see

Extensions that count your Claude messages became popular when the interface showed no counter. That gap has since closed, and they remain the category of tool where the permission you grant is most out of proportion to what you get back.

By the Continuum team. We build a workbench that runs Claude Code, Codex, and their peers, so the model rates quoted here are the ones our own cost analytics ship with.

The short version

Browser extensions observe the claude.ai page or its network traffic. They can count messages and surface the interface warnings, but they cannot see Claude Code, which is a terminal application whose traffic never touches the browser. Before installing one, check the first-party view: on a paid plan, claude.ai Settings then Usage shows progress bars for your 5-hour session and weekly limits, which is the number an extension is approximating. If you still want one, read its host permissions: access to claude.ai means access to everything you paste there.

What you need to know
  • Check Settings, then Usage on claude.ai first. It is the real number.
  • An extension sees claude.ai only. Claude Code never touches the browser.
  • Message counts are a proxy. Limits track tokens, not messages.
  • Read the host permissions: claude.ai access is access to everything you paste.
  • Anthropic ships a Chrome extension, but it is an agent, not a counter.

Check the built-in view first

The reason this category exists is that claude.ai used to show no usage figure at all. As of August 2026 it does, and it is authoritative in a way no extension can be.

The first-party answers, by surface.

You wantGo toYou get
Chat headroom on a paid planclaude.ai, Settings, then UsageProgress bars for the 5-hour session and weekly limits
Claude Code headroom/usage in the CLIThe same bars, plus an attribution breakdown
Claude Code headroom in an IDEThe VS Code extension Account and usage dialogThe same breakdown, Day and Week toggle
A team viewThe org analytics dashboardAdoption, sessions, accepted lines

What an extension can and cannot see

Coverage of a browser extension.

SurfaceVisible?Why
claude.ai chatYesIt is the page the extension runs on
Approaching-limit warningsYesRendered in the page
Message countsYesCountable in the DOM
Exact remaining quotaPartlyOnly what the page already reveals
Claude Code sessionsNoA terminal app; nothing passes through the browser
Claude DesktopNoNot a browser context
API key usageNoServer side, and not in your browser
Codex, Cursor, Gemini CLINoSame reason as Claude Code

The Anthropic extension is a different thing

Searching for a Claude Chrome extension surfaces an official one, and it is easy to install it expecting a counter. Claude in Chrome is an agent: it reads the page, clicks, types, fills forms, manages tab groups, and can read console output and network requests for debugging. As of August 2026 it is in beta on paid plans.

The permission question

An extension that reads your claude.ai session has read access to every conversation in it. For most developers that includes pasted source code, stack traces carrying internal hostnames, and occasionally a credential someone pasted and meant to remove.

  1. Check the host permissions. The manifest should declare claude.ai and nothing else. A match pattern like https://*/* is disqualifying for a message counter, whatever the listing says.
  2. Narrow it after installing. Open chrome://extensions, click Details, and set site access to run on specific sites or on click. A broad grant you never narrowed is a broad grant.
  3. Prefer open source with a repository you can actually read, and check that the published build corresponds to it rather than assuming.
  4. Check whether it phones home. A counter has no reason to contact a server. If it has an account system, ask what is being uploaded and where it lands.
  5. Watch for ownership changes. Small extensions with real install bases get bought, and the update that adds tracking is silent.

Why message counts mislead

Extensions count messages because messages are what the DOM contains. Your plan does not work that way: consumption tracks tokens, and it also moves with which model you chose and what effort level it is running at.

You sentExtension countsActual consumption
"What does this error mean?"1 messageNegligible
A 40-page PDF and a question1 messageSubstantial
Twenty short follow-ups20 messagesGrows each turn as context accumulates
One question on a high effort setting1 messageThinking tokens bill as output
A message after a two-hour gap1 messageCache expired; full context reprocessed

A counter can tell you that you have been busy. It cannot tell you how close you are. The third row is the one that trips people: twenty short messages cost far more than twenty times the first one, because the whole conversation is re-sent every turn.

Better answers by surface

What you want to trackUse
claude.ai chat headroomSettings, then Usage. An extension only if you want it in-tab
Claude Code headroom/usage
Claude Code spend historyccusage, or a desktop monitor
Both halves at once, continuouslyA monitor reading the rate-limit signal
A whole teamThe org analytics dashboard or the Analytics API

Questions people ask

Several third-party extensions exist for claude.ai in Chrome and Firefox. They count messages and surface the interface warnings. None can track Claude Code, which never passes through a browser, and on a paid plan the first-party Settings then Usage page already shows your limits.

No. Claude Code is a terminal application, so nothing it does reaches the browser. Use /usage inside the CLI, ccusage for spend history, or a desktop monitor for a continuous gauge.

On a Pro, Max, Team, or seat-based Enterprise plan, go to Settings then Usage. It shows progress bars for how much of your 5-hour session limit and your weekly limits you have consumed, and when each resets.

They need read access to claude.ai, which means access to everything in your conversations. Prefer open-source extensions whose manifest requests claude.ai only and that contact no server of their own, and narrow site access in chrome://extensions after installing.

No. Anthropic ships Claude in Chrome as a browsing agent that reads pages, clicks, types, and fills forms, in beta on paid plans as of August 2026. It is a much larger permission grant than a counter and it does not report your quota.

Because consumption tracks tokens rather than messages, and also moves with the model and effort level. A long document counts far more than a short question, and follow-ups in one conversation cost more each turn as context accumulates.

No. API usage is billed server side against a key and never renders in your browser. That belongs in the Console usage page, or in the Usage and Cost API.

A monitor that reads the rate-limit signal, since both draw on the same account allowance. Continuum does this without needing any browser access at all.

Sources

Every figure above was read from these pages on August 2026. Vendors reprice without notice; if you find a stale number, tell us.

  1. Claude help: usage limit best practices
  2. Claude for Chrome
  3. Chrome extensions: declare permissions
  4. Claude Code: manage costs effectively
Try it

No browser
access needed.

Continuum reads the rate-limit signal and the session files already on your machine, covering chat allowance and Claude Code in one gauge.

free app · your subscriptions · local-first