Claude Team analytics
and what they show per member.

How to open Claude Team analytics

Click your initials in the lower left of claude.ai and choose Analytics, or go straight to claude.ai/analytics/overview. On a Team plan, only Owners and the Primary Owner can open it. Admins can add members and change seat types, but they can't open Analytics, and neither can members.

Beside the overview there are pages for Chat, Claude Code, Cowork and Claude Design, and a Surveys page for asking members short questions. The numbers update once a day, so today's work shows up tomorrow. If someone's Claude Code work is missing, ask them to run claude update, because 2.0.28 is the first version that reports to the dashboard.

Members see only their own numbers. Settings, then Usage shows their limits and when they reset, and /usage in Claude Code adds the session's tokens and, with usage credits on, what they have spent this month against any limit set for them. The view that breaks a member's own usage down by product, model and skill is for usage-based Enterprise plans, and its switch doesn't appear on a Team plan.

What each Analytics page shows

The overview is built around four questions: who is using Claude, how, with what results, and at what cost. The product pages go a level deeper. Only some of the numbers break down by person, and the next section says which.

Can you see each member's usage on a Claude Team plan?

You can see some of it. The Claude Code page lists every member's accepted lines for the current calendar month, which you can search and export as a CSV, and the count starts again on the first of the month. With Anthropic's GitHub app installed, its leaderboard shows the top 10 members by pull requests or lines, each next to the part written with Claude Code, and Export all users downloads everyone else too. The Chat page names the top members by chats, projects and artifacts.

The spend report is the closest thing to a bill per person. In Settings, then Analytics, scroll to "How much is Claude costing?" and export it for this month, last month, the last 90 days or dates you pick within them. Each row is one person's use of one model, labelled with the product (Chat, Claude Code, Cowork or Office Agents), with the request count, prompt and completion tokens, and net and gross spend in dollars. The newest data is from yesterday, and the report only appears once usage credits are turned on.

What a Team plan doesn't have is a page per person, or an API to pull any of this into your own tools: Anthropic's Analytics API is for Enterprise. None of the pages the Help Center lists shows what anyone wrote to Claude either.

What does each member's usage cost?

We priced one engineer's 318 Claude Code sessions (September 4 to October 3, 2026) at Anthropic's API list prices. They used 67.9 billion tokens, 98.4% of them cache reads, and led to 601 merged pull requests. The total came to more than a hundred times the top of the monthly range Anthropic gives for an average developer. That engineer paid for Max 20x and usage credits, not a Team seat, but on a Team plan's spend report only the usage-credit part of a month like that would carry a price.

Members spend usage credits only once they pass their seat's limits, and Anthropic's cost docs say use inside the allowance isn't metered in dollars. The Value tab's costs per commit, pull request and session come from formulas printed on the tab, with inputs you can change, so check what spend a figure starts from before you quote it.

So we'd read Team's numbers for seat decisions, not for cost. Our engineer saw a limit notice on 16 of 29 active days, and a pattern like that, not a dollar figure, tells you who needs a Premium seat. The overview's cost section has a usage-limits view, and Organization settings, then Usage, lists each member's usage-credit spend this month, under Spend limits by user or, in Team organizations without per-member limits, Spend by org member.

What the Claude Code leaderboard measures

It mostly measures volume. In the 30 days to October 3, our heaviest user's laptop opened 596 of the 878 pull requests we merged, 68% of them, and the rest came from three other engineers, Codex and other machines. Ranked by pull requests or lines, that one person would have led by a distance. Anthropic's docs suggest using the leaderboard to find people others can learn from, and that is how we'd use it, not to rank engineers.

The counts are also low on purpose. Anthropic credits a merged pull request to Claude Code only when its lines match a session that edited the same files between 21 days before and 2 days after the merge, drops lines rewritten by more than 20%, and calls the result a deliberate underestimate. We count a different way, by the agent's signature on each pull request. Of the 2,620 pull requests merged in our two repos from July 5 to October 3, 2026, 95% carry an AI agent's signature and about 90% carry Claude Code's. How much code is written by AI explains why counts like these disagree, and since Anthropic's is built to come in low, read its share as a floor.

Those contribution numbers need repositories on github.com, the Claude GitHub app, and an Owner who turns on GitHub analytics under Organization settings, then Claude Code. They aren't available with zero data retention, and repositories on GitHub Enterprise Server get usage numbers only. Each merged pull request that qualifies gets a claude-code-assisted label, so you can find them on GitHub too.

How to get each member's tokens and cost on a Team plan

Turn on OpenTelemetry for everyone at once. Owners can put the telemetry variables in the env block of server-managed settings (Organization settings, then Claude Code, then Managed settings), and Claude Code fetches them at startup on each member's machine. Each metric then carries the member's email as user.email when Claude Code knows it, so claude_code.token.usage and claude_code.cost.usage group by person. The cost is a list-price estimate, and you run the collector and dashboards yourself; our OpenTelemetry setup for Claude Code has a config we tested.

These settings cover Claude Code's telemetry only; Cowork's is set up separately, and Chat isn't part of it. Zest records each Claude Code session through a plugin every engineer installs in about a minute (setup), prices it at list rates whether it fell inside the seat or not, and shows it next to the pull request it led to and the same engineer's Codex and Cursor sessions. It doesn't see Chat, Cowork or Claude Design, it connects to GitHub only, and its cost is an estimate, not your bill. Claude Code usage sets it beside Anthropic's other views.

{
  "env": {
    "CLAUDE_CODE_ENABLE_TELEMETRY": "1",
    "OTEL_METRICS_EXPORTER": "otlp",
    "OTEL_LOGS_EXPORTER": "otlp",
    "OTEL_EXPORTER_OTLP_PROTOCOL": "grpc",
    "OTEL_EXPORTER_OTLP_ENDPOINT": "http://collector.example.com:4317",
    "OTEL_EXPORTER_OTLP_HEADERS": "Authorization=Bearer example-token"
  }
}