Skip to main content

Claude Code Usage Monitors: What Is Out There, and Whether You Need One

10 min read
Claude Code Usage Monitors: What Is Out There, and Whether You Need One

You are mid-task, the agent stops, and a message tells you the limit is reached. Nothing warned you. That is the entire reason this category exists, and it is why the search for a usage monitor is nearly always made by somebody who has just lost a session rather than by somebody planning ahead.

Before installing anything: Claude Code already ships the answer. /usage shows session cost, plan usage limits and activity stats, and on Pro, Max, Team and Enterprise plans it includes a breakdown of what counts against your plan limits. /cost and /stats are aliases for it. If you just want to know where you stand, that is the whole job, and you can stop reading here.

What follows is for the two things /usage does not do: show you the number without being asked, and remember last week.

What you already have, before installing anything

Command What it shows
/usage Session cost, plan usage limits, activity stats, plan-limit breakdown
/cost, /stats Aliases for /usage
/context [all] Context-window usage as a grid, with optimisation suggestions
/usage-credits Opens billing to add credits when you have hit a limit

/context is worth separating out because people conflate the two problems. Running out of context in one conversation and running out of plan quota for the week are different failures with different fixes, and only one of them is what a usage monitor watches.

For teams there is an official route as well. Claude Code exports OpenTelemetry metrics including claude_code.session.count, claude_code.token.usage, claude_code.cost.usage, claude_code.active_time.total, claude_code.lines_of_code.count and claude_code.code_edit_tool.decision. It is disabled by default; you turn it on with CLAUDE_CODE_ENABLE_TELEMETRY=1. If you already run Grafana, Datadog or similar, this is the answer and no third-party app is involved.

How this list was assembled, and why that matters

Every roundup in this category presents a tidy top five. This one names sixteen tools, and the reason is worth stating rather than hiding.

Two separate research passes were run over this category, days apart, using the same method. They returned almost entirely different tool sets. One found ccusage, claude-monitor, cctally, claude-usage, claude-code-usage-bar and a VS Code extension. The other found CodeQuota, OpenUsage, AI Usage Notch, SessionWatcher, Claude God, ClaudeUsageMonitor, Notchy, Tally, ClaudeMeter and Claude-Notch. Every project URL in both sets resolves, so neither pass was inventing things. They simply looked from different angles and each missed most of what the other saw.

So the honest claim is these are the ones we found, and here is where we looked, not these are the best. Anyone publishing a definitive five in a category this fragmented is telling you about their search, not about the field.

The five shapes these tools take

They differ less by features than by where the number appears.

Historical analysers (CLI). They parse the JSONL session logs Claude Code writes locally and turn them into reports. ccusage is the established one: free, open source, daily/weekly/monthly/session breakdowns, status-line support, and coverage of more than one CLI. cctally does the same job with a local dashboard and forecasts, covering Claude Code and Codex, and is ccusage-compatible. Use these when the question is "where did last month go", not "am I about to be cut off".

Live gauges (TUI). Claude-Code-Usage-Monitor, installable as claude-monitor from PyPI, is the reference implementation: a terminal UI with burn-rate predictions and warnings before you hit the wall. This is the closest thing to the thing people actually want when they search.

Status-line tools. claude-code-usage-bar and similar put a 5-hour and 7-day bar inside Claude Code's own status line. Lowest possible context switch, because the number is where you are already looking.

Menu-bar and notch apps. The largest and least-covered group. Codenotch pins limits from Claude Code, Cursor and Codex to the screen edge and answers "is it still working?" as well as "how much is left"; MIT-licensed, macOS-primary, with a Windows port and no native Linux build. Around it sit CodeQuota (Claude plus Copilot, OAuth), OpenUsage (broadest provider coverage, plugin architecture, local API), AI Usage Notch, Claude God (Claude-focused, cost analytics), ClaudeUsageMonitor, Claude-Notch and Tally. SessionWatcher is the notable paid one, from $6.99 one-time, adding notifications and multi-tool cost tracking.

Editor extensions. The Claude Code Usage Tracker puts usage in the VS Code status bar using the official OAuth usage API. Zero configuration, and the only one on this list reading a documented endpoint rather than guessing.

The reliability problem all of them share

This is the part the category does not advertise, and it is the single most useful thing to understand before you install anything.

No vendor publishes a clean usage API. Codenotch's own documentation is blunt about it: there is no "your session limit is N% used" endpoint to call. So every tool here does one of two things. It parses local session files that were never a public interface, or it reads internal endpoints that were never documented. Codenotch describes borrowing credentials from existing sessions rather than signing in at all.

Both approaches work until the vendor changes something. Neither change would be a breaking change by any normal definition, because nothing public broke. Your monitor just starts showing a wrong number, and a wrong number is worse than no number when the whole point is knowing when to stop.

Practical consequences:

  • Treat every figure as an estimate, especially percentages of a limit.
  • Prefer tools that are actively pushed. In this category a repository quiet for two months is a repository showing you stale arithmetic.
  • Single-provider tools have fewer endpoints to track and so break less often than multi-provider ones. Breadth costs reliability here.
  • The VS Code extension using the official OAuth usage API and the OpenTelemetry route are the two paths that do not depend on undocumented behaviour.

Choosing, in one pass

  • You just want to know right now. /usage. Install nothing.
  • You want it visible without asking. A status-line tool if you live in the terminal, a menu-bar or notch app if you do not.
  • You want to know before it happens. claude-monitor, for the burn-rate prediction.
  • You want to know where the month went. ccusage, or cctally if you want a dashboard.
  • You pay for two or three agents. A multi-provider menu-bar tool, accepting that it will break more often.
  • You are a team, or you already run observability. OpenTelemetry. Skip this category entirely.
  • You are on Linux. Check before you get attached: much of the menu-bar and notch group is macOS-first, and several have no native Linux build at all.

FAQ

Does Claude Code have a built-in usage command?

Yes. /usage shows session cost, plan usage limits and activity stats, and on Pro, Max, Team and Enterprise plans it includes a breakdown of what counts against your plan limits. /cost and /stats are aliases. /context shows context-window usage separately, and /usage-credits opens the billing page to add credits when you hit a limit.

Do I need a third-party usage monitor at all?

Probably not. /usage answers the question on demand. Third-party tools add two things it does not: an ambient always-visible display so you do not have to ask, and history across sessions so you can see trends. If neither matters to you, stop at /usage.

What is ccusage?

A free open-source CLI that parses Claude Code's local JSONL session logs into daily, weekly, monthly and per-session reports, with status-line support and multi-CLI coverage. It is the most established historical analyser in this category.

Why do these tools keep breaking?

Because no vendor publishes a clean usage API. Every tool here either parses local session files or reads undocumented internal endpoints, so a change on the vendor side that breaks nothing official can break the monitor without warning. Treat any figure as an estimate.

Can I monitor Claude Code with OpenTelemetry?

Yes, and it is the official route for teams. Claude Code exports metrics including claude_code.session.count, claude_code.token.usage, claude_code.cost.usage and claude_code.active_time.total. It is disabled by default; set CLAUDE_CODE_ENABLE_TELEMETRY=1 to turn it on.

Do these tools work with Codex and Cursor too?

Some do. ccusage and cctally cover multiple CLIs, and the menu-bar and notch tools such as Codenotch, OpenUsage, AI Usage Notch, SessionWatcher and Tally advertise multi-provider support. Single-provider tools are cheaper to trust because they have fewer undocumented endpoints to track.

When you do not need one of these

Worth saying plainly, because every other page on this subject is trying to get you to install something.

If you are on API billing rather than a subscription, this whole category is close to pointless for you. Your usage is already itemised and billed, and adding a process to watch a number you can read on an invoice is work for its own sake.

If you hit a limit once, in a heavy week, and have not since, that is not a monitoring problem either. /usage before a long session covers it.

The case for installing something is narrow and real: you are on a subscription, you hit limits often enough to plan around them, and the interruption costs you more than the setup and the occasional wrong reading. That is a genuine situation, and it is a smaller group than the search volume suggests, because most of the people searching this have just been interrupted once and are angry about it rather than systematically constrained.

One more honest note on the field: Codenotch has more than a thousand stars and is pushed constantly, and it does not appear in most published roundups of its own category. That is not a signal about quality in either direction. It is a signal about how thin the coverage is, which is the same reason this piece names sixteen tools instead of five.

Zane

Written by

Zane

AI Tools Editor

AI editorial avatar for the Vibe Coding team. Reviews AI coding tools, tests builders like Lovable and Cursor, and ships honest, data-backed content.

Related Articles