Mixpanel Retention Analyzer

What it does

Mixpanel Retention Analyzer is a Level-2 MCP agent for product/growth analysts. It pulls retention and cohort reports from a read-only analytics API (Mixpanel or Amplitude) — or analyzes a provided export — and computes retention curves, per-period drop-off, cohort comparisons, and churn entirely deterministically in code. An LLM is used only to narrate insights, recommendations, and a summary over these code-derived numbers; it never produces or alters a metric. Ships with a deterministic template fallback when no LLM key or host sampling is available. Currently gated: without a sandbox analytics key, it runs against a provided export or deterministic mock fixtures with no live network calls. Ideal for teams who want trustworthy, reproducible retention diagnostics instead of LLM-hallucinated metrics.

Example prompts

• Run a full retention analysis for project test-project-001 for January 2026
• Analyze this retention export and show me drop-off and churn per cohort
• Compare cohorts in my export and tell me the best and worst performers
• Compute the retention curve for a single cohort with these retained-user counts

Tools (7)

Tools the agent exposes — your AI client calls them automatically when it needs them.

• run_full — Run the full retention analysis end-to-end: collect, compute curves/churn deterministically, narrate, and return the full JSON report.
• analyze_export — Deterministic analysis (no LLM) over a raw retention export: curves, drop-off, churn, findings, confidence.
• compute_retention_curve — Compute the deterministic retention rate curve, drop-off and churn for a single cohort.
• compare_cohorts — Compare cohorts in a retention export, ranking by final-bucket retention and churn.
• list_capabilities — List the agent's static capabilities, thresholds, providers, and limits.
• plan_inputs — Return questions, schema and a ready-to-edit example for a chosen tool.
• discover_intent — Co-design the exact input via clarifying questions before running.