Span Type — what kind of operation: mcp_tool = called a tool, llm_call = called a language model. Multiple rows with the same Trace ID = one conversation turn.
Trace ID — groups all steps of one conversation together. Click 🔍 on any row that has one to see the full execution path.
Destination — where the request was sent. mcp:// = tool call, http:// = LLM or internal.
Latency — wall-clock time from when the request was sent to when the response came back, including network.
Tokens — how many words (roughly) were sent to and returned from the LLM. Higher = more expensive.
Cost — estimated USD spend for this single call based on the model's pricing. Zero for MCP/tool calls.