GA4 Site Blood Test

What it does

GA4 Site Blood Test runs a deterministic "blood test" on your website using Google Analytics 4. It collects key engagement and conversion metrics by device (mobile / desktop / tablet) — bounce rate, average session duration, session conversion rate, pageviews per session, and engagement rate — compares them against industry reference ranges, computes an overall health score from 0 to 100, and surfaces critical and warning findings (including a derived mobile-vs-desktop gap signal). A deterministic core derives every number. An LLM (Anthropic or OpenAI) only narrates the prescription on the synthesis step and is never allowed to invent or change metrics; without an API key it degrades gracefully to a template prescription. A rule-based recommendation engine then matches each finding to real B2B SaaS products and surfaces ranked vendor candidates. Gated: GA4 access requires a Google Cloud service account with Viewer access to the property (GA4_PROPERTY_ID + service-account JSON); the LLM step requires an Anthropic or OpenAI key. Both degrade gracefully. Tools: - run_health_check — full end-to-end blood test: collect, diagnose, score 0–100, prescribe, recommend vendors. Returns AnalysisResult. - get_metrics_panel — raw GA4 health panel by device. - get_worst_pages — top-N worst pages by bounce rate. - analyze_findings — pure deterministic diagnosis of a HealthPanel (no LLM). - get_prescription — LLM prescription from findings (template fallback). - list_available_metrics / get_metrics_thresholds — tracked metrics with healthy/warning/critical bands and direction. - search_vendor_catalog / get_vendor_details / match_to_finding — B2B SaaS vendor catalog search and finding-to-vendor matching. - plan_inputs — interview helper returning questions, JSON schema, and a ready-to-edit example for a tool. - ping — health check. Array/object arguments may be passed as JSON or 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.