Claude Prometheus Exporter

A Prometheus exporter for Claude API usage and cost metrics.

Features

• Token Metrics: Track input, output, cache read, and cache creation tokens
• Request Metrics: Monitor API request counts
• Cost Metrics: Track API costs in USD
• Automatic Collection: Continuous metric collection with configurable intervals
• Simple: Single-file implementation following the same pattern as updownio-exporter

Prerequisites

• Python 3.13+
• Poetry (for dependency management)
• Claude Admin API key (starts with sk-ant-admin...)

Installation

poetry install

Configuration

Configuration via environment variables or config files:

Environment Variables

export APIKEY="sk-ant-admin-..."
export PROMETHEUS_PORT=9001
export LOOP_INTERVAL=30
export LOGGING_LEVEL=INFO

Config Files

Place a JSON config file at:

• /etc/claude-prometheus/claude-prometheus.json
• ~/.config/claude-prometheus.json

Example config:

{
  "apikey": "sk-ant-admin-...",
  "prometheus": {
    "port": 9001,
    "namespace": "claude"
  },
  "loop": {
    "interval": 30
  },
  "logging": {
    "level": "INFO"
  }
}

Configuration Options

Option	Description	Default
apikey	Claude Admin API key	Required
prometheus.port	Port to expose metrics on	9001
prometheus.namespace	Prometheus namespace	claude
loop.interval	Seconds between collections	30
logging.level	Log level (DEBUG, INFO, WARNING, ERROR)	INFO

Note: The exporter always fetches data from midnight (UTC) to now on each collection. It uses delta calculation to avoid double-counting, so counters only increment with new usage since the last collection.

Usage

poetry run python claude_prometheus.py

Or with custom config:

export APIKEY="sk-ant-admin-..."
export LOOP_INTERVAL=60
poetry run python claude_prometheus.py

Metrics Endpoint

Once running, metrics are available at:

http://localhost:9001/metrics

Exported Metrics

All metrics use the claude_ namespace prefix (configurable).

Token Metrics

• claude_tokens_total (Counter): Total tokens consumed by type
  • Labels: type (input, output, cache_read, cache_creation), model, api_key_id, api_key_name, owner_email, owner_name
• claude_tokens_current (Gauge): Tokens in current period
  • Labels: type, model, api_key_id, api_key_name, owner_email, owner_name

Request Metrics

• claude_requests_total (Counter): Total API requests
  • Labels: model, api_key_id, api_key_name, owner_email, owner_name
• claude_requests_current (Gauge): Requests in current period
  • Labels: model, api_key_id, api_key_name, owner_email, owner_name

Cost Metrics

• claude_cost_total_usd (Counter): Total cumulative cost in USD
  • Labels: workspace_id
• claude_cost_current_usd (Gauge): Cost in current period in USD
  • Labels: workspace_id

Daemon Metrics

Provided by daemon-metrics:

• claude_prometheus_loop_duration_seconds: Duration of each collection loop
• claude_prometheus_loop_success: Whether the last loop succeeded
• claude_prometheus_item_success: Per-API-key success metrics

Getting an Admin API Key

1. Log in to the Claude Console
2. Navigate to the Admin settings (requires admin role)
3. Generate an Admin API key

Note: Standard API keys (starting with sk-ant-api) will NOT work.

Example Prometheus Configuration

Add this to your prometheus.yml:

scrape_configs:
  - job_name: 'claude-exporter'
    scrape_interval: 60s
    static_configs:
      - targets: ['localhost:9001']

Example Queries

Total tokens by model

sum by (model) (rate(claude_tokens_total[5m]))

Total tokens by API key owner

sum by (owner_name, owner_email) (rate(claude_tokens_total[5m]))

Total tokens by API key

sum by (api_key_id, api_key_name) (rate(claude_tokens_total[5m]))

Request rate by owner

sum by (owner_name) (rate(claude_requests_total[5m]))

Top users by token usage

topk(10, sum by (owner_name, owner_email) (claude_tokens_current))

Top API keys by usage

topk(10, sum by (api_key_name, owner_name) (claude_tokens_current))

Current cost by workspace

sum by (workspace_id) (claude_cost_current_usd)

Development

Code Quality Tools

# Format code
poetry run black claude_prometheus.py

# Type checking
poetry run mypy claude_prometheus.py

# Linting
poetry run pylint claude_prometheus.py
poetry run flake8 claude_prometheus.py
poetry run pycodestyle claude_prometheus.py

Architecture

Following the same pattern as updownio-exporter, this is a single-file implementation with:

• Configuration management via the-conf
• Daemon metrics via daemon-metrics
• Clean collection loop with proper timing
• Prometheus metrics via prometheus_client

Troubleshooting

Authentication Errors

Ensure you're using an Admin API key (starts with sk-ant-admin).

Permission Errors

Your API key must have admin permissions to access usage/cost data.

No Data in Metrics

• Check that your organization has recent API usage
• Adjust loop.lookback to a longer period
• Check logs for API errors

License

MIT