What it does
Funnel Cohort Analyst computes funnel step conversion (overall, step-over-step, and drop-off) plus a cohort retention matrix from event data you pass in as arguments. It is account-free: there is no warehouse connector and no required credentials — you hand it the events and it does the math. A pure deterministic core derives every number; an LLM narrative layer only describes those numbers and is never allowed to invent or change them. With no API key and no sampling-capable host, it falls back to a deterministic template, so the numeric output is always reliable. Guardrails: event data is sanitized and treated as data (prompt-injection defanged), user IDs are aggregated for counting only and never echoed back (no PII leak), only aggregated numbers reach the LLM, and runs are idempotent. Tools: - run_full — full pipeline: deterministic funnel + cohort math, then LLM narrative (template fallback). - analyze_funnel — deterministic funnel + cohort math only, no LLM. - list_capabilities — static discovery of tools and analysis parameters. - plan_inputs — interview helper that returns the questions, JSON schema, and a ready-to-edit example for a tool. Inputs accept array/object arguments as JSON or as a JSON string. Missing required inputs return a structured needs_input payload (questions + schema + example) instead of erroring, and upgrade to a native elicitation form on clients that support it.
