Claude for Teams and Enterprise organisations use claude.ai/analytics/claude-code, open to Admins and Owners, which adds GitHub-backed contribution metrics and a leaderboard. Console (API) organisations use platform.claude.com/claude-code, open to anyone with the UsageView permission, which adds per-member spend. Neither has a repository dimension, neither shows remaining quota, and neither covers Claude Code running on a cloud provider. Verified August 2026.
- Two dashboards, two URLs. Your sign-in method picks one, not your preference.
- Contribution metrics need a GitHub app and an Owner to enable them.
- Attribution is deliberately conservative. Read the counts as a floor.
- Zero Data Retention orgs get usage metrics only, no contribution data.
- Neither dashboard has a repository dimension or live quota.
Which dashboard your organisation gets
The two first-party dashboards, as of August 2026.
| Your plan | URL | Who can open it | What it adds |
|---|---|---|---|
| Claude for Teams or Enterprise | claude.ai/analytics/claude-code | Admins and Owners | Contribution metrics from GitHub, leaderboard, CSV export of all users |
| Claude Console (API) | platform.claude.com/claude-code | UsageView: Developer, Billing, Admin, Owner, Primary Owner | Daily spend, and spend plus accepted lines per member this month |
| Bedrock, Google Cloud, Microsoft Foundry | None | n/a | Nothing. That traffic is not reported to Anthropic |
What the Teams and Enterprise dashboard shows
The summary tiles.
| Metric | Definition |
|---|---|
| PRs with CC | Merged pull requests containing at least one line written with Claude Code |
| Lines of code with CC | Effective lines across merged PRs attributed to Claude Code |
| PRs with Claude Code (%) | Share of all merged PRs that contain attributed code |
| Suggestion accept rate | Share of Edit, Write, and NotebookEdit proposals accepted |
| Lines of code accepted | Lines accepted in session, excluding rejections and later deletions |
Below the tiles are four charts: Adoption (daily active users and sessions), PRs per user (merged PRs divided by daily active users), a Pull requests breakdown splitting merged PRs with and without Claude Code, and a Leaderboard of the top ten contributors. The leaderboard has an Export all users button that downloads the full contribution data as CSV, not just the ten shown.
Turning on contribution metrics
Usage and adoption data is on by default for every Teams and Enterprise account. The PR and lines-of-code half requires connecting GitHub, which is a four-step job split across two admins.
A GitHub admin installs the app
Install the Claude GitHub app on your organisation at github.com/apps/claude. GitHub Cloud and GitHub Enterprise Server are both supported.
A Claude Owner enables Claude Code analytics
Go to claude.ai/admin-settings/claude-code. The Owner role is required; an Admin cannot do this step.
Enable the GitHub analytics toggle
Same page, second switch. This is the one people miss, and the dashboard then reports "GitHub app required" forever.
Authenticate and pick organisations
Complete the GitHub flow and select which GitHub organisations to analyse. Data typically appears within 24 hours and updates daily.
How PR attribution actually works
This is the part worth understanding before anyone quotes the number in a board deck, because the algorithm is public and it is deliberately biased low.
- Added lines are extracted from the merged PR diff.
- Claude Code sessions that edited matching files inside the window are identified.
- PR lines are matched against session output by several strategies, after normalising whitespace, collapsing runs of spaces, standardising quotes, and lowercasing.
- Metrics are computed for attributed lines against total lines, and the PR is labelled
claude-code-assistedin GitHub.
The rules that decide what does not count.
| Rule | Effect |
|---|---|
| Window: 21 days before merge to 2 days after | Work older than three weeks is invisible to attribution |
| Rewritten by more than 20% | Not attributed to Claude Code at all |
| Lock files, generated code, build directories, fixtures | Excluded |
| Lines over 1,000 characters | Excluded as minified or generated |
| Branch of origin | Ignored; the algorithm does not consider source or target branch |
Pulling the data programmatically
Which API matches which dashboard.
| If your dashboard is | Use | Key |
|---|---|---|
platform.claude.com/claude-code | Claude Code Analytics API | Admin API key, sk-ant-admin01-... |
claude.ai/analytics/claude-code on Enterprise | Claude Enterprise Analytics API | Analytics API key with read:analytics |
claude.ai/analytics/claude-code on Teams | No API. Export the spend report CSV | n/a |
The Enterprise Analytics API key is created by the primary owner only, at claude.ai/admin-settings/api-access, after enabling public API access. Its endpoints live under /v1/organizations/analytics/ and it is rate limited at the organisation level, 60 requests per minute across all of them by default.
Reporting without wrecking trust
This is the part that goes wrong most often, and it is not a technical problem. A leaderboard of named engineers ranked by AI usage is a policy decision that ships enabled.
- Announce collection before you start. Not after someone finds the dashboard.
- Report team aggregates. Individual comparison changes behaviour immediately, and not in the direction you want.
- Never rank people by usage. High usage is not skill, and low usage is often judgement.
- Do not tie it to performance. The moment you do, you are measuring compliance rather than adoption.
- Say what it is for. "We are deciding whether to renew" is a reason people accept.
- Watch acceptance rate as a tool signal. Measuring a person on it rewards whoever reviews least carefully.
The dimension neither dashboard has
No org-level source knows which repository the work happened in, and repository is the most actionable cut there is, because an expensive repo is usually a fixable repo rather than a badly behaved engineer. Nor does any of them expose remaining quota, so an admin cannot see who is about to be blocked.
Both dimensions exist only locally: the working directory is recorded on every request in the session transcripts, and quota headroom arrives in rate-limit response headers on live traffic.
Questions people ask
Yes, two. Claude for Teams and Enterprise organisations use claude.ai/analytics/claude-code, open to Admins and Owners. Console customers use platform.claude.com/claude-code, open to anyone with the UsageView permission.
Because your engineers are probably on subscription seats rather than API keys. Usage inside a seat allowance is not metered in dollars, so the billing view is empty while the activity is real. Use the claude.ai analytics dashboard, or the spend report once usage credits are turned on.
A GitHub admin installs the Claude GitHub app at github.com/apps/claude, then a Claude Owner enables Claude Code analytics and the GitHub analytics toggle at claude.ai/admin-settings/claude-code and completes the GitHub authentication. Data typically appears within 24 hours.
Three common causes: the GitHub app is not installed, the separate GitHub analytics toggle was never switched on, or your organisation has Zero Data Retention enabled, in which case contribution metrics are not available at all and only usage metrics appear.
Not from any org-level source. Repository is only recorded locally, in the session transcripts, which capture the working directory for each request. It has to come from a tool running on the machine doing the work.
There is no benchmark worth quoting. Watch your own trend: a plateau usually indicates a specific setup or workflow obstacle for the non-adopting group rather than a preference, and the fix is removing the obstacle rather than promoting the tool harder.
No, and Anthropic says so. Contribution metrics count only high-confidence attributed lines and are described as an underestimate, and Console spend figures are labelled estimates for analytics purposes. Use the billing page for money.
No. Quota state is reported in rate-limit response headers on live requests and is not exposed by any dashboard or organisation endpoint.
Sources
Every figure above was read from these pages on August 2026. Vendors reprice without notice; if you find a stale number, tell us.