Publish PostHog MCP Analytics dashboards and alerts

Goal

Turn the starter's canonical $mcp_tool_call events into a reusable, privacy-safe learning loop for activation, tool success, latency, failure, undo, feedback, handoff, and return usage.

Scope

• Dashboard and saved-insight definitions.
• First-inbound to first-value funnel.
• Safe breakdowns for tool/intent category, new/returning, model/provider version, and test/cohort flags.
• Latency and abnormal-failure monitors.
• Privacy audit and test-traffic exclusions.

Acceptance criteria

• A documented import/setup path creates the core dashboard and alerts.
• Raw message, feedback, identifier, value, URL, token, tool input/result, and provider error data remain excluded.
• A privacy test samples event properties and fails on prohibited shapes.
• The template documents a recurring review-to-issue workflow.