The best Claude Code token tracking tools

Honest roundup for developers who want to know where their Claude tokens go. Updated April 2026.

Why look for a Claude Code token tool?

Anthropic's dashboard tells you that you spent money. It doesn't show where it went. The /usage command surfaces only the current 5-hour rate-limit window; /cost returns "using subscription" for Pro/Max users. If you're on a Max plan or running heavy agentic workflows on the API, you need a tool that breaks down spend by session, project, detector trigger, or prompt — not just aggregate quota.

What to look for

Live vs retrospective
Does it watch sessions in real time, or only summarize after the fact?
Cross-client scope
Claude Code only, or does it also cover Cursor, VS Code, Copilot, JetBrains?
Local-first vs cloud
Does your session data stay on your machine, or does it get uploaded?
Surface
CLI, statusline, GUI, or inside Claude Code as an MCP / skill?
Cleanup included
Just visibility, or does it also remove stale MCP servers, memories, and skills?
Pricing model
Free / open source, one-time, subscription, or enterprise?

The tools

Naqi

Free (scan + 7-day snapshot) · Pro $59/yr or $7.99/mo · 14-day refund

The live token watchdog + AI workspace auditor

Naqi watches every Claude Code session in real time with 9 waste-pattern detectors, sends a weekly TokenDiet digest, and also scans MCP servers, memories, and skills across 10 AI clients — with backup + diff + undo on every cleanup.

Strengths

Only tool in this list that combines live session monitoring with cross-client config cleanup. 10 AI clients supported. Native 10MB desktop app (macOS + Windows + Linux). Local-first scanning, zero telemetry.

Limitations

Pro features (live watcher, 9 detectors, weekly digest, batch cleanup) are paid ($59/yr or $7.99/mo). Free tier shows a 7-day Token Hygiene snapshot but not the live watcher.

Best for: Daily Claude Code power users on Max plans who want live alerts before mega-sessions blow up the bill, and who also juggle MCP servers across multiple clients.

getnaqi.com →

CCM (Claude Code Management)

Free · open source

Bash CLI with usage commands

CCM by @dr5hn is a bash CLI tool that parses the JSONL session files Claude Code writes to `~/.claude/projects/`. Commands include `ccm usage sessions`, `ccm usage history --days 30`, `ccm usage dashboard`, and `ccm session summary`.

Strengths

Lightweight, zero dependencies, native terminal workflow, open source. Perfect if you live in the shell and want pipeable output. Estimates cost with model-specific pricing.

Limitations

CLI only — no GUI, no live watcher, no cross-client scope. Claude Code only. No cleanup features.

Best for: Terminal-first developers who already script their Claude Code workflow and want token visibility without leaving the shell.

github.com/dr5hn/ccm →

claude-lens

Free · open source

Statusline burn-rate indicator

claude-lens by @Astro-Han renders live usage percentage, pace delta (are you burning faster or slower than the reset window?), and countdown to reset directly in your Claude Code statusline — using the `rate_limits` data Anthropic ships in statusline stdin since v2.1.80.

Strengths

Zero-switch UX — live data appears inline while you work. No external app to check. Open source. Great for awareness without interruption.

Limitations

Real-time only, no historical aggregation. Statusline only. Claude Code only. No cleanup features.

Best for: Developers who want ambient token awareness without context-switching to a separate dashboard.

github.com/Astro-Han/claude-lens →

Session-cost MCP server

Free · open source

Cost tools as MCP, stays inside your AI client

An MCP server (by @smigolsmigol) that reads local session data and exposes 5 cost tools directly inside Claude Code, Cline, and Cursor: session cost, cumulative per-project costs, cache savings analysis, monthly projection vs Max subscription, and subagent cost attribution.

Strengths

Lives inside your AI client — ask Claude "what did I spend on this project?" and get a real answer. Auto-detects installed AI coding tools. Can run as a SessionEnd hook for automatic cost logging.

Limitations

MCP-surface only — no standalone UI. Requires MCP setup. Claude Code / Cline / Cursor only.

Best for: Developers already heavy on MCP who want cost questions answered in-flow rather than in a separate dashboard.

GitHub issue #33978 (original post) →

CodexBar

Freemium · paid multi-provider

Broad multi-provider usage dashboard for macOS

CodexBar is a native macOS (Swift) app that tracks AI usage quotas across 16 providers — OpenAI, Anthropic, Google, xAI, and more. Freemium with paid multi-provider.

Strengths

Polished macOS-native UI. Broadest provider coverage in this list. Good if you span Claude + ChatGPT + Gemini + Grok at once.

Limitations

Aggregate quota tracking, not per-session attribution. No cleanup features. macOS only. Not Claude Code-specific — strategy is breadth over depth.

Best for: Multi-provider users who want a single menu-bar dashboard across all the AI services they pay for.

codexbar.app →

Anthropic's native /usage and /cost

Free · included with Claude Code

Built into Claude Code, limited visibility

The `/usage` command shows your current 5-hour rate-limit window and weekly quota. `/cost` returns per-session input/output tokens but says "using subscription" for Pro/Max plans (no price calculation). The Claude Code statusline added rate-limit fields in v2.1.80.

Strengths

Zero setup, already there. Officially supported. Accurate for API-billing users.

Limitations

No historical aggregation, no waste-pattern detection, no cross-session analysis, no per-project breakdown, and /cost is crippled for subscription users. See GitHub issue #33978 for the 10+ open requests asking Anthropic to ship more.

Best for: API-pay-as-you-go users who just want session totals. Not enough for Max-plan users trying to understand quota drain.

See /usage feature gap — GitHub #33978 →

At a glance

Tool	Live watcher	Historical	Cleanup	Cross-client	Pricing
Naqi	✓ (Pro)	30-day (Pro)	✓ across 10 clients	10 clients	Free (scan + 7-day snapshot) · Pro $59/yr or $7.99/mo · 14-day refund
CCM (Claude Code Management)	—	30-day	—	Claude Code only	Free · open source
claude-lens	✓ (statusline)	—	—	Claude Code only	Free · open source
Session-cost MCP server	—	Per-project	—	Claude Code, Cline, Cursor	Free · open source
CodexBar	—	Per-provider	—	16 providers	Freemium · paid multi-provider
Anthropic's native /usage and /cost	—	Current window only	—	Claude Code only	Free · included with Claude Code

Which should you pick?

I want live alerts before a mega-session blows up my bill

→ Naqi Pro — the live watcher + 9 detectors are the only setup in this list that fires before the damage is done.

I live in the terminal and just want pipeable session data

→ CCM — fits a CLI-first workflow with zero extra UI.

I want ambient awareness without leaving Claude Code

→ claude-lens (statusline) or the session-cost MCP server (in-flow). Both are free.

I care about Claude + OpenAI + Gemini + Grok quotas in one place

→ CodexBar — broadest provider coverage.

I also have MCP sprawl across Cursor, VS Code, Copilot, JetBrains

→ Naqi — the only tool in this list that also audits MCP servers, memories, and skills across 10 AI clients.

I only want official, zero-setup tooling

→ Anthropic's native /usage and /cost commands — limited, but free and built in.

Try them all

Most of these are free or have free tiers. The honest answer is: try the ones that fit your surface (CLI, statusline, GUI, in-app MCP) and keep the ones that stay useful after a week. Naqi's free tier is designed as a genuinely useful tool, not a teaser — scan once and decide.

Start with a free scan

Free forever. No account. 10 AI clients. 3 seconds.

Download for macOS

Last updated: April 2026