The best Claude usage trackers, ranked and tested

The one question that decides it

Claude usage data comes from two places that have nothing to do with each other, and a tool can only report what its source contains.

The shortlist

Before you install anything: /usage

Most roundups skip this, and it is the single best-value option because it is already on your machine and it is the only local tool with a server-side view of your plan bars.

/usage

# press d  last 24 hours
# press w  last 7 days
# press r  retry if the usage endpoint was rate limited

On a Pro, Max, Team, or Enterprise plan it shows plan usage bars plus an attribution breakdown: recent usage split across skills, subagents, plugins, and individual MCP servers, each as a percentage of the total. It also raises behavior flags for things like long context or cache misses when one accounts for 10% or more of recent usage. That attribution is genuinely hard to get any other way, and it is how you find the MCP server quietly eating a fifth of your allowance.

ccusage: the reference for history

ccusage by @ryoppippi parses the JSONL transcripts your agents already write and turns them into a cost report. No install, no configuration, works retroactively on data you have had for months, and it has become the number other tools are checked against.

npx ccusage@latest daily        # by date
npx ccusage@latest weekly       # by week
npx ccusage@latest monthly      # by month
npx ccusage@latest session      # by conversation
npx ccusage@latest blocks       # by 5-hour block

# useful flags
npx ccusage@latest daily --breakdown --since 20260701
npx ccusage@latest daily --instances          # group by project
npx ccusage@latest daily --json               # pipe it somewhere

It no longer reads only Claude. As of August 2026 it parses local logs from Claude Code, Codex, OpenCode, Amp, Droid, Codebuff, Goose, Kilo, Kimi, Qwen, GitHub Copilot CLI, and Gemini CLI among others, which makes it the cheapest way to answer "what did all my agents cost this month".

Claude Code Usage Monitor: the terminal live view

The best-known community TUI. It watches the local session logs, derives a burn rate, and predicts when you will hit the window, with realtime, daily, and monthly views. MIT licensed.

uv tool install claude-monitor      # recommended
# or: pipx install claude-monitor
# or: pip install claude-monitor

claude-monitor                     # aliases: ccmonitor, ccm
claude-monitor --plan max5         # pro | max5 | max20 | custom

The honest boundary: prediction from local signals is not the same as reading your account. It is very good at "at this burn rate you have about forty minutes", and it is estimating rather than reporting the ceiling it measures against.

Continuum: both halves, every agent

We built Continuum because nothing above covered both halves across more than one agent. It reads the live rate-limit signals for 5-hour and weekly gauges with a burn-rate estimate of time remaining, and separately parses the session files for history, using the same deduplication rule ccusage established and pricing each event at the rate in effect when it happened.

• Live gauges for the 5-hour and weekly windows, with time to exhaustion rather than a bare percentage.
• Multi-account. Several Claude and Codex subscriptions side by side, each with its own meter, so you can start the next session on the one with headroom.
• Ambient. Mac menu bar, web, Windows and Linux desktop, iPhone, and watch. You do not run a command to find out.
• Cross-agent. Claude Code and Codex CLI in one view, plus the other agents it drives.
• Per-repo attribution that survives git worktrees, so one project does not fragment into six rows.

Statusline, extensions, and the org-wide options

Statusline

The least intrusive option of all: render usage into the Claude Code prompt itself so it is in front of you on every turn without a second window. ccusage ships a statusline subcommand for this, and Claude Code can display context-window usage in the status line directly.

Browser extensions

Extensions count messages on claude.ai by observing the page or its network traffic. They are the only option for someone who never opens a terminal, and they come with two real caveats.

• They cannot see Claude Code. It is a terminal program and never touches the browser, so an extension is blind to the half of your allowance an agent is consuming.
• The permission is broad. You are granting page access on a site where you paste code, credentials, and internal documents. Read what it asks for before installing, and prefer one whose source you can read.

OpenTelemetry and the Analytics API

For an organisation the local tools are the wrong layer entirely. OpenTelemetry export streams per-user token and cost metrics from each developer machine into your own observability stack in near real time, and it is the only option that works regardless of whether people are on subscriptions, the Console, or a cloud provider. Alongside it: the Console dashboard and Claude Code Analytics API for API organisations, and the spend report CSV plus Enterprise Analytics API on seat-based plans.

How to tell a tracker is honest

Two trackers reading the same files routinely disagree, and the reasons are always one of four things. These are worth knowing whether you are choosing a tool or writing one.

Questions people ask

Keep reading