Skip to main content

Monitoring and Observability

AI Gateway captures telemetry data as metrics and events for every MCP method call it handles. Configure exporters to choose how and where you want to see that data.

telemetry:
  metrics:
    - mcp.tools.call
  exporters:
    stdout_logs_exporter:
      type: stdout

Telemetry Data Types

Metrics

Metrics cover every MCP method the gateway handles, such as initialize, tools/list, and tools/call, dimensioned by attributes like toolkit, tool, and outcome. Use the zilla metrics command to see the raw metrics that AI Gateway has collected.

Logs

Logs signal important activities and collect relevant details with them. They are most useful for troubleshooting issues like authorization failures or upstream connection problems, and they also add observability around the MCP request lifecycle. The easiest way to see logged events is to add the stdout exporter to your zilla.yaml config.

Exporters

Exporters determine how the collected telemetry data is exposed. The Prometheus exporter exposes any configured metrics on the standard /metrics endpoint that Prometheus scrapes. The stdout exporter logs all events to the host stdout, where any log aggregator can collect them. The OTLP, AWS CloudWatch, and Syslog exporters push logs and metrics to their respective external systems.