Docs/Task-Level Cost Tracking

Task-Level Cost Tracking

Track costs at the MCP task level — not just per-request — to understand the true cost of agent workflows.

What is task-level tracking?

A single agent task (e.g. "refactor the auth module") may involve dozens of individual tool calls: reading files, running searches, editing code, executing tests. SatGate groups these calls by task ID so you can see the total cost, duration, and tool breakdown for each logical unit of work — not just isolated API hits.

How it works

Task IDs are automatically detected from upstream MCP server responses. When an MCP server includes a task identifier in its protocol messages, SatGate captures it and correlates all subsequent tool calls within that task context.

  • SatGate intercepts MCP protocol messages flowing through the proxy
  • Task boundaries are detected from upstream server metadata
  • Each tool call is tagged with its parent task ID
  • Costs are aggregated at both the per-call and per-task level

Viewing task costs

Navigate to the Tasks dashboard to see a table of all tracked tasks. For each task you'll see:

  • Task ID — the unique identifier from the MCP server
  • Total cost — sum of all tool call costs within the task
  • Duration — time from first to last tool call
  • Tool breakdown — which tools were called and how much each cost
  • Agent — which agent token executed the task

Prerequisites

Requirements:

Without per-tool costs configured, tasks will still be tracked but cost values will show as zero. Task duration and tool call counts are always available regardless of cost configuration.

Use cases

  • Cost attribution: "This coding task cost 450 credits across 38 tool calls over 12 minutes."
  • Efficiency analysis: Compare task costs across agents to identify which are most efficient.
  • Budget planning: Use historical task costs to set realistic per-agent budgets.
  • Anomaly detection: Spot tasks with unusually high tool call counts or costs.

Related