Per-developer usage reporting: how to get a Claude Code number for each person

Per-developer usage reporting for Claude Code, Anthropic's coding tool, comes from a different source on each of the three setups, and only one source works on all three. Anthropic's documentation on managing costs, read on 4 October 2026, names them: a spend report file and an Enterprise Analytics API on Teams and Enterprise plans, a dashboard and a Claude Code Analytics API on the Claude Console, and OpenTelemetry export or a gateway on a cloud provider [1]. This page explains each in plain words and says which to use. It is part of the guide to Claude Code costs for teams, and the screens that show totals, as opposed to per-person figures, are on the page on where to see Claude Code spend.

Three terms appear throughout. A token is a piece of text the model reads or writes, and all usage is counted in tokens. A CSV file is a plain spreadsheet file, one row per record, which opens in Excel or Google Sheets. An API is an interface that lets a program, instead of a person, ask a service for data; an API key is the secret string that proves which account the program belongs to. A dashboard is a web page of figures and charts. A cloud provider means Amazon Bedrock, Google Cloud's Agent Platform or Microsoft Foundry. A gateway is a server the company runs between Claude Code and the model provider. OpenTelemetry is explained in its own section below.

Teams and Enterprise plans: the spend report file and the Enterprise Analytics API

On a Teams or Enterprise plan, usage draws from each member's seat allowance, the usage included in their seat, and Anthropic says usage inside the allowance "isn't metered in dollars" [1]. Per-person dollar figures therefore describe usage credits, the paid usage a member draws on after the allowance runs out.

The spend report CSV. Anthropic's documentation says that on the Teams plan an administrator exports the spend report CSV, "which lists token usage and estimated spend per user and per model" [1]. The report is updated daily and appears once usage credits are turned on [1]. This is the one per-person file available on the Teams plan, and it answers the two questions a monthly review asks: who spent what, and on which model.

The analytics dashboard export. The dashboard at claude.ai/analytics/claude-code has an Export all users button that downloads contribution data for every user as a CSV, including users outside the top 10 shown on the leaderboard [2]. Contribution data means pull requests and lines of code shipped with Claude Code assistance, which needs the GitHub integration an Owner enables [2]. It is a measure of output per person, which pairs with the spend report's measure of cost per person.

The Enterprise Analytics API. On the Enterprise plan, Anthropic's documentation says this API "returns per-user usage and cost reports across Claude surfaces, including Claude Code" [1]. Two conditions apply. A Primary Owner creates the key, with the read:analytics scope, at claude.ai/analytics/api-keys [1]; a scope is the list of things a key is allowed to do. And, in the words of the analytics documentation, "The API is not available on the Teams plan" [2]. Reveneau's reading: on Enterprise, this is the source to build a monthly report from, because it covers chat, Cowork and Claude Code in one place; on Teams, the spend report CSV is the source.

The Claude Console: the dashboard and the Claude Code Analytics API

The Claude Console is Anthropic's web console for accounts billed per token. Anthropic's documentation names two per-person sources on it.

The Console dashboard at platform.claude.com/claude-code has a team insights table with one row per user who has authenticated to Claude Code [2]. Each row shows spend this month, the user's total API costs for the current month, and lines this month, their accepted code lines [2]. Users who signed in with an API key appear by key identifier, and users who signed in with an account login appear by email address [2]. Opening the dashboard needs the UsageView permission, granted to the Developer, Billing, Admin, Owner and Primary Owner roles [2]. Anthropic notes that the spend figures are estimates for analytics purposes and that actual costs are on the billing page [2].

The Claude Code Analytics API returns, in Anthropic's words, "the same daily per-user metrics programmatically with an Admin API key" [1]. The Admin API key is created at platform.claude.com/settings/admin-keys [1]. Programmatically means a script fetches the figures instead of a person reading the page, which is what a finance team needs to match the figures to a cost centre list each month.

Cloud providers: nothing returns to Anthropic, so the data has to be collected

On Amazon Bedrock, Google Cloud's Agent Platform and Microsoft Foundry, which are the Amazon, Google and Microsoft services that offer Claude models inside a company's cloud account, Anthropic's documentation says Claude Code "does not send metrics from your cloud back to Anthropic", so the dashboards and the Claude Code Analytics API above do not cover this usage [1]. The cloud billing console shows the total. For per-person figures, Anthropic names three options: OpenTelemetry, a Claude apps gateway, or an LLM gateway [1].

OpenTelemetry: the one method that works on every setup

OpenTelemetry is an open standard for sending measurements and event records from a program to a monitoring system. In Claude Code's case, each developer's copy of the tool sends its figures to a system the company runs. Anthropic's cost documentation says OpenTelemetry export "works on every setup and is the only option that streams per-user token and cost metrics into your own observability stack in near real time" [1]. An observability stack is the company's monitoring system, such as the tools an operations team already uses to watch servers.

Anthropic's monitoring documentation, read on 4 October 2026, says it is turned on with environment variables, which are named values a program reads at start-up: CLAUDE_CODE_ENABLE_TELEMETRY=1 turns export on, OTEL_METRICS_EXPORTER=otlp picks the standard export format, and OTEL_EXPORTER_OTLP_ENDPOINT gives the address of the receiving system [3]. The default export interval is 60,000 milliseconds, which is one minute [3]. A developer's own setting can still switch a signal off or redirect it to the console, so Anthropic says to set the exporter selectors in managed settings too "if you need them locked", and marks that behaviour as requiring Claude Code version 2.1.223 or later [3]. Managed settings are settings the organisation deploys to every machine that a developer cannot override.

Claude Code exports eight metrics, listed in the monitoring documentation [3]:

Metric name	What it counts	Unit
claude_code.session.count	Sessions started	none
claude_code.lines_of_code.count	Lines of code modified	none
claude_code.pull_request.count	Pull requests created	none
claude_code.commit.count	Git commits created	none
claude_code.cost.usage	Cost of the session	USD
claude_code.token.usage	Tokens used	tokens
claude_code.code_edit_tool.decision	Permission decisions on code edits	none
claude_code.active_time.total	Active time	seconds

The two that matter for a budget are the cost counter and the token counter. Each is incremented after every API request, and each includes the model name, a query_source of main, subagent or auxiliary that separates the main conversation from helper requests, and the effort level [3]. The token counter also includes a type of input, output, cacheRead or cacheCreation, so a report can show how much of a developer's usage was read from the prompt cache at the lower price [3].

The per-person part comes from the attributes attached to every metric. Anthropic's list includes user.account_uuid and user.account_id, which identify the account; user.email, the sign-in email, included when available; organization.id; and session.id [3]. A company can add its own labels through the OTEL_RESOURCE_ATTRIBUTES variable, and Anthropic's example is department=engineering,team.id=platform,cost_center=eng-123, which it says allows filtering by team or department and tracking costs per cost centre [3]. Anthropic adds a warning: each custom key becomes a label on every metric series, so values with many distinct entries increase storage cost in the monitoring system [3].

Two limits apply. Anthropic's note reads: "Cost metrics are approximations. For official billing data, refer to your API provider (Claude Console, Amazon Bedrock, or Google Cloud's Agent Platform)" [3]. And by default the cost figure is at list price; the modelPricing managed setting, explained on the page on where to see Claude Code spend, changes the exported cost metric to contracted rates. The sister guide's page on how to check your cache hit rate shows what the token counter's cache types reveal.

Gateways: per-person attribution through one shared credential

A gateway is a server the company runs between Claude Code and the model provider. Anthropic's gateway overview, read on 4 October 2026, says developers authenticate to the gateway with their own credential while the gateway forwards every request with one company credential, "so authentication, usage tracking, budgets, and audit logging happen in one place you control" [5]. Per-person reporting follows from the developer credential: Anthropic's rollout advice is to "issue each developer a gateway credential, so usage is attributed to the developer" [4].

A Claude apps gateway is Anthropic's own self-hosted gateway, included in the claude program. Anthropic says it provides, on the cloud providers, "per-user usage attribution, OTLP metrics with token counts, and per-user spend limits" [1]. Developers sign in through /login with the company's identity provider, the system the company already uses to sign staff in to other services, and the gateway emits OpenTelemetry usage metrics to the company's monitoring system [5]. Its spend-limits documentation adds a reporting endpoint, /effective, that returns each person's resolved cap and period-to-date spend, with a sort option of spend_desc that lists the top spenders first [6]. The same page says that by default the provider's bill attributes everything to the one shared credential, and that on Amazon Bedrock a per-developer AWS cost attribution option changes this [6]. One limit: the sign-in is a browser step that a person approves, and there is no sign-in method for a program running alone, so an automated pipeline with nobody present cannot authenticate through it and is configured against the provider directly [5].

An LLM gateway is a gateway product the company already runs. Anthropic's cost documentation describes the option as routing all Claude Code traffic "through a proxy that tracks spend per key", and states that "Several large enterprises reported using LiteLLM, an open-source tool that tracks spend by key. This project is unaffiliated with Anthropic and has not been audited for security" [1]. That is the whole of what Anthropic says about it, and this page adds nothing to it. The gateway documentation adds that Anthropic "doesn't endorse, maintain, or audit third-party gateway products", and that a gateway maintained separately needs its forwarding rules updated as Claude Code changes [4].

Every method in one table

Method	Plans it works on	What you get, per person
Spend report CSV	Teams and Enterprise	Token usage and estimated usage-credit spend per user and per model, updated daily [1]
Analytics dashboard export	Teams and Enterprise, with GitHub integration	Pull requests and lines of code with Claude Code assistance, all users [2]
Enterprise Analytics API	Enterprise only	Usage and cost reports across Claude chat, Cowork and Claude Code; key with read:analytics scope from a Primary Owner [1][2]
Console dashboard, team insights	Claude Console	Spend this month and lines this month per member; estimates [2]
Claude Code Analytics API	Claude Console	The same daily per-user metrics, fetched by a script with an Admin API key [1]
OpenTelemetry export	Every setup	Token and cost counters per user, model, request source and token type, one minute apart by default; approximations [1][3]
Claude apps gateway	Cloud providers and the Anthropic API	Per-user attribution, OpenTelemetry metrics, per-user spend limits, a top-spenders listing [1][6]
LLM gateway	Any provider the gateway forwards to	Spend per developer key; Anthropic names LiteLLM as reported in use, unaffiliated and unaudited [1]

Which method to use

Reveneau's recommendation. On the Teams plan, export the spend report CSV monthly and match it against the analytics export, so each person has a cost and an output figure. On the Enterprise plan, produce the monthly report from the Enterprise Analytics API, because it covers every Claude surface in one call. On the Claude Console, use the Claude Code Analytics API for the same reason. On a cloud provider, set up OpenTelemetry export first, because it is the only method Anthropic says works everywhere and the only one that reports during the month; add a Claude apps gateway when the company also wants per-person spend limits, which the page on spend limits for an organisation, a group and a person explains. Whatever the method, reconcile the per-person figures to the invoice once a month, because every per-person figure in this table is an estimate by Anthropic's own description.

Reveneau is an AI software development consultancy. All of its code is written by AI, so token use is a running cost of every Reveneau build, and Reveneau is independent of Anthropic. Every API name, scope name and metric name on this page is from Anthropic's documentation read on 4 October 2026. If you want help setting up per-developer reporting, contact Reveneau.