Monitor activity

Once agents are connected and making tool calls, you can monitor everything from the dashboard.

Dashboard

The dashboard groups key metrics into themed cards:

• Gateway health — the share of tool calls that executed successfully, with a count of real errors (timeouts, upstream failures; policy blocks excluded)
• Inventory — counts of agents, users, resources, and policies
• LLM usage — spend, tokens, and distinct models across your agents
• Gateway traffic — total requests through the gateway and how many were blocked

Below the cards, a tool executions trend sits beside an LLM usage cost-over-time chart. The bottom row shows the top models by token usage, recent activity, and top blocked tools. The LLM widgets populate once agents report usage through OpenTelemetry.

Audit log

Navigate to Audit Log to see a real-time feed of every action. Each entry shows the timestamp, actor (user or agent), the resource and tool involved, and the outcome.

Events include tool executions, agent connections, policy changes, resource changes, and connection events. Click any entry to expand and see full details including metadata like IP address and browser.

Use the filter bar to narrow the feed by action type, agent instance, agent (for example, all Claude Code clients), or resource.

Analytics

For aggregated trends over time — tool usage, LLM token spend and cost, and policy-enforcement metrics — open Analytics. It has three tabs: MCP Usage, LLM Usage, and Security.

See the Analytics guide for a full breakdown of each tab, and OpenTelemetry for how to feed the LLM Usage tab with token and cost data from Claude and Codex.