Tool reference

Start with list_projects for ids, and get_insights for open-ended questions — it returns conclusions with evidence, corrected for the traps the raw tools warn about.

Analysis

Tool	Answers
get_project_summary	Headline numbers against the prior window.
get_event_trend	A time series for direction over time.
list_top_events	Which events fire most.
break_down_events	Splits by country, plan, signup_source, and more.
analyse_funnel	Ordered conversion between events, sequenced per user.
get_retention	Cohort retention by first-seen period.
get_user_distribution	How activity spreads across users.
compare_cohorts	Whether doing X predicts doing Y, activity-adjusted.
list_persons / get_person	Recently active users and one user's history.

Money and acquisition

Tool	Answers
get_revenue	What was paid, by whom, by source and store.
get_subscriptions	The lifecycle: base, movement, trials, churn, refunds, realized LTV.
get_marketing_performance	Impressions, installs, and spend by channel.
compare_channels	Acquisition against activation and revenue, by channel.

The fix-verify pair

Tool	Does
create_annotation	Drops a release marker at a point in time. Needs mcp:annotate.
compare_window	The same window before and after a marker: totals, error counts, per-event deltas. Flags an after-window that has not fully elapsed, so an undercount is never read as a drop.

Two habits the tools themselves will remind you of: use analyse_funnel for any conversion question (dividing event counts overweights heavy users), and treat an empty revenue result as "nothing connected", never as "no sales".