Claude Code managed settings for cost tracking

Why use managed settings for Claude Code telemetry?

• Every machine reports. Shell profiles depend on each developer remembering, and a missing developer looks the same as one who spent nothing.
• Developers can't override it. Managed settings sit above command-line, project and user settings. When they set an OTEL_EXPORTER_OTLP_* variable, Claude Code removes conflicting developer-set variables at startup.
• Repositories can't do it. Claude Code ignores the OpenTelemetry exporter variables in a repository's .claude/settings.json, so committing them to your repos won't turn telemetry on.

Anthropic's monitoring docs name managed settings as the way administrators configure telemetry for all users.

Which settings turn on Claude Code cost tracking?

Seven environment variables, set in the env block. Replace the key with one from Profile & setup in Ganivra.

{
  "env": {
    "CLAUDE_CODE_ENABLE_TELEMETRY": "1",
    "OTEL_LOGS_EXPORTER": "otlp",
    "OTEL_METRICS_EXPORTER": "none",
    "OTEL_EXPORTER_OTLP_PROTOCOL": "http/protobuf",
    "OTEL_EXPORTER_OTLP_ENDPOINT": "https://api.ganivra.com/v1/otlp",
    "OTEL_EXPORTER_OTLP_HEADERS": "Authorization=Bearer gv_live_...",
    "OTEL_LOGS_EXPORT_INTERVAL": "60000"
  }
}

Use one workspace key for the whole company and tell teams apart with attributes, so all coding-agent spend lands in one workspace. Anyone with access to a machine can read the file, so treat the key like any other deployed secret, and replace it everywhere if it leaks.

Where do Claude Code managed settings go?

Pick the delivery method your IT team already uses. Each one carries the same JSON.

The file-based option also reads every *.json file in a managed-settings.d directory next to managed-settings.json, merged in alphabetical order. Admin-console settings need a Claude Team or Enterprise plan. Details: managed settings and server-managed settings.

Which managed settings source wins?

When a machine gets settings from more than one method, Claude Code by default uses only the highest-ranked source that delivers any setting and ignores the rest. From highest to lowest:

1. The Claude admin console (server-managed settings)
2. Device management: the macOS profile or the HKLM registry key
3. The managed-settings.json file and managed-settings.d directory
4. The HKCU registry key on Windows

The common mistake: if someone already set any policy in the admin console, a managed-settings.json file you deploy with telemetry is ignored entirely. Either put the telemetry variables in the same source, or switch Claude Code to merge mode, which needs Claude Code 2.1.242 or later. Within the env block, a managed value overrides the same variable in user and project settings (settings precedence).

How do I label each team's Claude Code spend?

Add OTEL_RESOURCE_ATTRIBUTES. Ganivra groups spend by team.id, team or cost_center, in that order, and keeps other keys as filters. With files, deploy the shared settings above everywhere and a small per-team file such as managed-settings.d/50-team.json:

{
  "env": {
    "OTEL_RESOURCE_ATTRIBUTES": "team.id=payments,cost_center=eng-142"
  }
}

With device management, use one profile per team group instead. Follow the format rules from Anthropic's docs:

• Comma-separated key=value pairs, with no spaces: team.id=data-platform, or percent-encode, as in team.id=data%20platform.
• Quotes don't escape spaces: team.id="Data Platform" keeps the quotes in the value.
• Developers with no team attribute appear under No team in Ganivra, so gaps are easy to spot.

Will Ganivra know which developer made each request?

It depends on how developers sign in:

• Claude Team, Enterprise or another Claude account: each event carries the developer's email, and Ganivra shows spend under it. No extra setup.
• An API key, Amazon Bedrock, Google Vertex AI or Microsoft Foundry: there's no Claude account in the session, so Claude Code sends only a random per-install ID. Spend still adds up by team, but developers appear as IDs rather than names.

For the second group, add ganivra.user_id to each machine's attributes, for example from your device-management tool's per-user variables or a login script that writes the per-machine file:

"OTEL_RESOURCE_ATTRIBUTES": "team.id=payments,ganivra.user_id=priya.n"

What do developers' machines send?

With the settings above, each model request sends its model, input, output, cache-read and cache-write tokens, timing, session and the developer's identity. Prompt text, responses and tool details are redacted by default and stay redacted unless someone sets OTEL_LOG_USER_PROMPTS, OTEL_LOG_ASSISTANT_RESPONSES or OTEL_LOG_TOOL_DETAILS. Leave all three unset: cost tracking doesn't need them, and Ganivra reads only the usage fields. Telling developers this up front avoids most rollout pushback.

How do I check every developer is reporting?

1. 01
   Test one machine

   Run any prompt in Claude Code. Within a minute or two, the request appears in Ganivra under Executions with an ID starting cc-.

2. 02
   Check names and teams

   Open the Coding agents tab and group by developer, then by team. Look for No team or unnamed IDs.

3. 03
   Compare with your headcount

   After a week, compare the developers reporting with the people who have Claude Code seats or keys. Anyone missing either hasn't used it or isn't covered by the policy.

4. 04
   Debug a silent machine

   Start Claude Code with claude --debug-file <path> and look for [3P telemetry] errors. Check the protocol is http/protobuf, not gRPC, and that no higher-ranked settings source is hiding your file.

One exception to know: in sessions started from the Claude desktop app, a variable already set in the launch environment takes precedence over settings files.

What does a safe rollout look like?

1. Pilot: five developers on one team for a few days. Confirm names, teams and costs look right.
2. One department: deploy the shared file and per-team files through device management. Fix any No team rows.
3. Everyone: roll out company-wide and announce what is and isn't collected.
4. Use the data: after two to four weeks, use cost per developer at API rates to decide between Team seat tiers and API billing. See Claude Code Team seats vs API.

For the single-developer version of this setup, see how to track Claude Code cost per developer, and the integration guide for every option.