Codeburn Open source

What it does

Codeburn is a local-first dashboard and CLI that tracks how much your AI coding tools are actually costing you. It reads the session files that tools like Claude Code, Cursor, Codex, and GitHub Copilot already write to disk, then breaks down token usage and spend by model, project, task, and tool — with no account, no API keys, and nothing uploaded anywhere.

What is inside

• An interactive terminal dashboard with daily/weekly/monthly views, plus a browser dashboard (codeburn web, served locally)
• codeburn optimize: finds waste (re-read files, low edit ratios, unused MCP servers, bloated configs) and can auto-apply fixes with undo
• codeburn compare: side-by-side model performance/cost comparisons
• codeburn yield: correlates spend against shipped Git commits/PRs
• codeburn guard: soft/hard budget caps with session warnings
• codeburn quota: shows subscription-plan usage (Claude Pro/Max, Cursor Pro, Copilot) against limits
• Desktop apps (macOS/Windows/Linux), a macOS/Windows menu bar app, a GNOME Shell extension, and an MCP server mode for Claude Code, Cursor, and other MCP clients
• JSON/CSV export, multi-currency support (162 ISO codes via ECB rates), and device pairing to merge usage across machines

Works with

Codeburn tracks roughly 40 AI coding tools and agents out of the box, including Claude Code, Cursor, Codex, Cline, Gemini CLI, GitHub Copilot, Grok, Roo Code, OpenCode, and Zed. It also exposes itself as an MCP server that Claude Code and Cursor can call directly.

How to install or connect

npx codeburn

For a persistent command, npm install -g codeburn; a codeburn menubar mode is available on macOS/Windows, and native installers (.msi, .deb, .rpm, AppImage, or Homebrew) exist per platform. Requires Node.js 22.13+ and at least one supported tool with session data already on disk.

Maintenance and safety

MIT-licensed, actively maintained, and popular (11k+ GitHub stars, reportedly used by 150k+ developers). Its core privacy claim is that everything runs locally — no wrapper, no proxy, no API keys, nothing leaves your machine — and optional desktop telemetry defaults off in the EU/EEA/UK and is opt-in elsewhere, sending only anonymous usage buckets, never code.

Who should use it

Anyone running one or more AI coding CLIs/agents who wants visibility into where their token spend and subscription quota are actually going, and to catch config waste before it costs real money. It's a companion/observability tool, not a coding agent itself.

Pricing: completely free and open source, with no paid tier.