Show HN: Claude Code Usage Monitor – real-time tracker to dodge usage cut-offs

Original Article ↗ Hacker News Discussion ↗

Installation & Packaging

  • Multiple commenters want an easier, self-contained install: ideally a single executable or a proper Python package installable via uv, pipx, etc.
  • Current setup requires a globally installed Node CLI (ccusage) plus Python; some see this Python requirement as a mismatch given Claude Code is a Node tool.
  • Others note uv tool install avoids duplicating Python, and that a more standard project structure (e.g., pyproject.toml) would simplify one-line installs.

How Usage Monitoring Works

  • The tool reads Claude Code’s verbose logs in ~/.claude/projects/*/*.jsonl, which contain full conversation history plus metadata.
  • It targets fixed-cost Claude plans (Max x5/x10/etc.), not API pay-per-use.
  • Planned features include:
    • “Auto mode” using DuckDB and ML to infer actual token limits per user instead of hardcoded numbers.
    • Exporting usage data (e.g., per-project) and exposing cache read/write metrics via flags.

Pain Points with Claude Code Limits & Billing

  • Several users want a simple command that just shows, “how much of my plan is used,” and clearer separation between subscription and API credits.
  • Confusion is common around Claude vs Anthropic billing UIs and which actions consume API credits (e.g., GitHub integration unexpectedly spending from the API wallet).
  • Some report extremely high implied API usage values (thousands of dollars) on flat-rate Max plans and speculate about margins vs losses.
  • Experiences with limits differ: some hit them quickly when scanning large codebases; others run long Opus sessions without issues. The exact Pro/Max token limits remain unclear and disputed.
  • One user notes token usage seemingly doesn’t reset after a window unless 100% is reached, which feels punishing.

Auth & Login UX

  • Strong dislike for email “magic link” / no-password logins; seen as tedious, easy to abandon, and harmful to active usage.
  • Others argue email-based flows are actually more secure and simpler for non-technical users who constantly reset passwords.

Feature Requests & Ecosystem

  • Requests for similar tools for Cursor and Gemini, and for making this monitor callable directly as a Claude tool.
  • People share related tools: cursor usage monitors, multi-session UIs for Claude Code, and Datadog/OTel-based monitoring.

Code Quality & “Vibe Coding” Debate

  • Some criticize the project as mostly a thin wrapper around ccusage, with a large monolithic Python file, hardcoded values, and emoji-heavy README, reading as “vibe-coded.”
  • Others defend the informal style for a free hobby tool and argue that if it works and surfaces useful metrics, that’s acceptable.

Energy / CO₂ Tracking Tangent

  • A semi-serious request appears for estimating power/CO₂ per session based on tokens; this prompts:
    • Jokes about “low-carbon developers” and carbon-tiered AI plans.
    • Skepticism about the practical value of per-token CO₂ metrics, given aviation/industry dwarfs such emissions.
    • A broader debate on the effectiveness of individual conservation efforts vs systemic contributors.