Anthropic Launches CC-Ledger: Cost Tracking Dashboard for Claude Code Sessions
Key Takeaways
- ▸CC-Ledger provides real-time cost tracking and attribution for AI coding sessions at the PR level, answering what merged PRs actually cost
- ▸Session cost distribution metrics (p50, p95, p99) help engineering leaders identify runaway spend before it shows up on invoices
- ▸Auto-classification of agent turns into planning, coding, debugging, and review helps leaders understand whether spending is exploratory or productive
Summary
Anthropic has introduced CC-Ledger, an open-source tool that provides engineering leaders with real-time visibility into the costs of AI-assisted coding sessions using Claude Code, Cursor, and other AI coding assistants. The tool captures every prompt, tool call, and code change into a local SQLite database and surfaces three critical metrics: the true cost of merged PRs, identification of "runaway burn" sessions, and breakdown of where AI spend goes—toward planning and exploration versus productive coding.
CC-Ledger answers a growing operational need as AI coding assistants become standard. The dashboard aggregates token spend per merged PR (with cache reads discounted at 10%) and provides session-cost distribution visibility at p50, p95, and p99 percentiles to surface cost outliers before they impact invoices. The tool also auto-classifies agent turns into planning, coding, debugging, and review categories, enabling leaders to distinguish exploratory from execution spend.
Installation is minimal: a single command wires CC-Ledger's lifecycle hooks into Claude Code with no external daemons or SaaS infrastructure. All telemetry remains local, with optional GitHub App integration committing session records to a private git branch. The dashboard supports drill-down by user, repository, and time period, giving directors and VPs the visibility they need to justify and optimize AI spend.
- One-command installation with fully local execution (SQLite, no external SaaS) means teams can adopt cost visibility without new dependencies or privacy concerns
Editorial Opinion
CC-Ledger addresses a real blind spot for engineering teams adopting AI coding tools at scale. By localizing all telemetry and providing drill-down cost visibility without external SaaS infrastructure, Anthropic has built something leaders can actually use to justify AI spend and identify which teams and workflows get the most leverage. As AI development costs rise, this kind of operational transparency will become table-stakes.
