Track Claude Code costs across your team

Per-session cost for every engineer, rolled up by person, team, model and kind of work

Start free for up to 5 engineers. Team views come with Pro.

To track Claude Code costs across a team, price each session from its own tokens and attach it to the engineer, repo and kind of work it belongs to. VibeCheck, Maestro AI's coding agent visibility product, does this for Claude Code, Codex and the Cursor Desktop Agent: a stop hook hands each finished session to a background sync service, and Usage rolls the cost up by person, team, model, kind of work and skill.

The Problem

The provider console tells you what the organization spent and who spent it. That answers "how much" and "who", but not "on what". Fernhollow Labs, a fictional 40-engineer team, rolled Claude Code out to everyone in one sprint. A month later the platform lead could see a total and each person's share, but not which team, which kind of work, or which sessions drove it. Two engineers also used Codex, and their spend lived in a different console.

How It Cascades

Per-key totals can't tell a feature from a debugging loop, so the expensive work stays invisible.

Each tool reports separately, so nobody sees the team's whole agent spend in one place.

Without per-session cost, the only lever is a blunt cap, and the engineers getting the most from agents hit it first.

The spreadsheet someone builds from exports goes stale the week after it's made.

The Insight

The session is the unit that explains the bill. It has an engineer, a repo, a model, a kind of work and, often, a pull request. Price it from its own tokens and every other view is a roll-up.

The Solution

Each engineer runs one install command. In Claude Code, a stop hook nudges a small background service and returns at once; it never waits on the network. That service uploads each finished session and retries if the laptop is offline. VibeCheck prices every session from its own tokens, cache included, at current published reference prices, and labels the figures as estimates. Usage shows Tokens or Cost over 4, 8 or 12 weeks or weeks you pick, broken down by person, team, model, kind of work and skill. Sessions sorts by cost, so the most expensive work is one click away. Connect the GitHub App and Usage adds cost per authored PR.

The Outcome

The platform lead at our fictional team opens Usage on a Monday and sees which teams and kinds of work drove last month's Claude Code spend, with Codex beside it, and the five most expensive sessions to look at first.

Questions

No. The Claude Code stop hook nudges a small background service and returns at once; it never waits on the network. The service uploads each finished session and retries if the laptop is offline.

No, and they're labeled as estimates. Each session is priced from its own tokens, cache included, at current published reference prices. Negotiated discounts and subscription plans aren't applied, so the figures are built for comparing people, teams, models and kinds of work.

Yes. Sessions are captured on each engineer's machine, so Bedrock-routed and Anthropic-direct sessions are both supported. Cost is estimated at published reference prices, not your Bedrock rates.

See your team's Claude Code spend, session by session

Price every session, then roll the spend up by person, team, model and kind of work.

Start free for up to 5 engineers. Team views come with Pro.

After you sign up

  1. 1InstallThrough your MDM, or one command per laptop.
  2. 2SessionsEach one appears as it finishes, priced and read.
  3. 3Weekly RewindOn Pro, the Monday after an engineer’s first week.