Claude Code Metrics

Metrics groundcover collects from Claude Code.

The metrics groundcover collects from Claude Code via its built-in OpenTelemetry export. Each metric is listed under its exact name: search or query by that name, or link directly to its anchor.

Anthropic's monitoring reference documents these metrics with OTel dot names (for example claude_code.token.usage); groundcover stores them in the Prometheus form shown here (dots become underscores, with the unit and _total suffixes appended).

Labels are the dimensions you can slice these metrics by. Every metric carries the standard set: session_id, organization_id, user_id, user_email, user_account_id, user_account_uuid, service_name, service_version (the Claude Code version), os_type, os_version, host_arch, terminal_type, claude_deployment_mode. Metrics with additional labels:

Label

Description

On metrics

Example values

type

Breakdown dimension

token usage (token type), lines of code, active time

input, output, cacheRead, cacheCreation / added, removed

model

Model that served the request

token usage, cost

claude-sonnet-4-6

query_source

Where the request originated

token usage, cost

agent_name, skill_name, plugin_name, mcp_server_name, mcp_tool_name, effort

Active subagent / skill / plugin / MCP context when applicable

token usage, cost

start_type

How the session started

session count

fresh, resume, continue

decision

Permission decision

code-edit-tool decisions

accept, reject

tool_name

Editing tool the decision applies to

code-edit-tool decisions

Edit, Write, NotebookEdit

language

Language of the edited file

code-edit-tool decisions

source

What triggered the decision

code-edit-tool decisions