What problem does it solve?
When an AI agent connects to your PostHog MCP server, you need to understand what it actually did — which tools it called, in what order, what it was trying to accomplish, and where it went wrong. This Skill turns raw $mcp_tool_call events into a readable per-session narrative so you can debug failing agent runs and audit who is connecting to your MCP.
Core Features & Use Cases
- Session listing: List recent MCP sessions with tool call counts, duration, client name, and resolved person identity, with search and sorting support.
- Tool-call replay: Read one session's tool calls chronologically, including intent, duration, and error messages, to reconstruct the agent's run.
- Goal summarization: Generate and cache an LLM summary of a session's goal from its recorded
$mcp_intent values.
- SQL fallbacks: Use HogQL recipes for errored sessions, effective (unwrapped) tool names, and cross-session aggregation when the typed tools can't answer the question.
- Use Case: A user pastes an MCP analytics sessions URL and asks "what did this session do?" — you list the session, pull its tool calls with the correct
date_from, and generate an intent summary explaining the agent's goal and where it failed.
Quick Start
Show me the tool calls and goal for MCP session <session_id>, using its session_start as the date_from parameter.
