Triple Whale

Run a root-cause analysis on a change in the store's metrics — "why did my ROAS drop last week?", "what caused the spike in CAC?". Returns whether it's a real anomaly, a summary of the likely cause(s), and findings with supporting evidence and suggested actions. Use this for "why did X change / drop / spike" questions instead of trying to diagnose it with raw SQL. This can take longer than other tools.

explain-metric-change

Search Triple Whale's library of verified ClickHouse SQL examples by natural-language question (semantic search over global and shop-specific examples). ALWAYS call this before drafting new SQL. When a returned example closely matches the request, adapt it with minimal changes instead of writing new logic — the examples encode correct table choices, metric definitions, and ClickHouse syntax patterns.

find-sql-examples

Get the analysis context for your connected store(s): store id and name, currency, IANA timezone, default attribution model and window, start-of-week day, industry, GMV segment, sales platform, pixel status, and today's date in the shop timezone. All money is in each store's currency and all dates are in its timezone. Call this once at the start of every data conversation. If you are connected to a single store, call it without arguments. If you are connected to several stores, pass shopIds for the store(s) you are analyzing (omit to get the context for every connected store, returned as one entry per store). The other Triple Whale data tools and your SQL depend on these values.

get-shop-info

Get the store's configured topline KPIs from the Triple Whale Summary page — Blended ROAS, MER, Net Profit, New-Customer CPA, Total Sales, Blended Ad Spend, etc. — exactly as the shop has them configured (custom formulas, manual costs and assumptions included). Prefer this over run-sql for topline/overview metrics: SQL cannot reproduce these configured values. Pass a date range; add compareStartDate/compareEndDate for a period-over-period comparison.

get-summary-kpis

Fetch the exact column definitions (names, types, required filter fields, descriptions) for named warehouse tables, formatted for writing ClickHouse SQL. Call this for every table you plan to reference before writing SQL. Always include the required fields noted in the schema in your WHERE clause. If the returned schema is empty, the table is not valid for this shop — pick another table from get-available-tables.

get-table-schemas

List the Triple Whale store(s) this connection can access, each with its status: mcpEnabled (whether the store can be queried here) and multiShopEnabled (whether it can be combined with other stores). Use the returned shopId values EXACTLY as shopId/shopIds in the other tools. Call this when you are unsure which stores are available, or whenever a tool returns a permission / "not on an MCP plan" / "not one of your connected stores" error — it confirms the exact valid store domains you can work on instead of guessing.

list-stores

List the warehouse tables available for this store, with a description of each, optionally filtered by business domain (store_business, order, customer, product, subscription, marketing_entity, session, customer_journey, attribution, operations, customer_experience). Call this BEFORE get-table-schemas to validate which tables exist. The response includes disambiguation guidance for commonly-confused tables (e.g. email_sms_settings_table for campaign content/templates vs email_sms_table for time-series performance metrics).

get-available-tables

Get attributed performance (revenue, conversions, ROAS) broken down by channel/source using Triple Whale's pixel attribution engine and the chosen attribution model — distinct from raw ad-platform reported numbers. Use this for attribution questions ("revenue by channel under Triple Attribution", "first-click vs last-click by source"). For platform-reported delivery metrics, use run-sql on ads_table instead.

pixel-attribution

Resolve a relative period ("last week", "this month", "30 days ago") into exact inclusive ISO dates in the shop's timezone. Use the timezone and startOfWeekDay from get-shop-info. Semantics: "last N days" = rolling completed-day window ending yesterday; "this X" = start of the period through today; "N X ago" = one full past calendar period; "all_time" returns startDate=null. Use the returned dates in your SQL event_date filters. Reuse dates already resolved this conversation instead of calling again.

get-date-range

Execute a read-only ClickHouse SELECT query against the store's warehouse and return the data as JSON: { columns, rows, totalRows, truncated, stats, dateWarning }. Results are capped (default 200 rows, max 500; the cap is enforced inside the query) — when truncated is true, totalRows is null and more rows exist: refine the query with GROUP BY or LIMIT instead of paging. Every table must have an explicit event_date filter, otherwise data may silently default to the last 7 days (a dateWarning is returned when no date filters were applied). On a SQL error, fix the query using the schemas you fetched and retry (at most twice). This tool never modifies store or warehouse data.

run-sql