MCP App Store
by KeenEthics
Power My Analytics icon

Power My Analytics MCP Server for ChatGPT

By Power My Analytics

Overview

Connect your AI assistant to your marketing and e-commerce data across 80+ platforms — including Facebook Ads, Google Ads, Shopify, Stripe, TikTok, LinkedIn, and more. Once connected, you can ask questions in plain English and get answers pulled directly from your live data, without writing a single query. Capabilities: - Compare performance across platforms - Detect anomalies and trends - Query custom date ranges - Create and manage datasets - Get account health summaries

How to connect

  1. ChatGPT: Open the Power My Analytics listing in the ChatGPT app directory and choose Connect. ChatGPT listing Power My Analytics exposes 35 tools here.

Tools

kb_get_article

ChatGPT
Returns the full published markdown of one article. Use the path from kb_list_articles or kb_search, e.g. "public/application/sources/facebook-ads.md".

kb_list_articles

ChatGPT
Lists every published knowledge-base article (path, title, category) from the live mirror. Drafts still in review are open PRs and are not listed.

kb_list_categories

ChatGPT
Lists the category paths (folders) that exist in the published knowledge base.

kb_validate_article

ChatGPT
Checks a draft against the KB structural rules (single title, heading hierarchy, "What does this mean for me?" under every limitation, required connector-guide sections). Errors block kb_create_article; kb_update_article blocks only on errors the change introduces (the published article's existing errors are listed, not blocking). Run this first.

pma_add_data_table

ChatGPT
Add a new data table to an existing dataset. A data table is a saved query for one connector account — it defines which platform, which account, which fields, and which date range to pull data from. Each dataset can hold multiple data tables (one per platform or account), and their data is blended together when you call pma_get_dataset_data. Use this when the user wants to add a new data source or report to an existing dataset. Use pma_list_datasets to find a valid dataset_id. Use pma_list_data_sources to find valid connector, account_id, and sub_account_ids values.

pma_analyze_cohorts

ChatGPT
Builds monthly retention cohorts: customers grouped by the month they FIRST appear, then tracked for how many come back in later months — "of January’s new customers, how many returned in February? March?" USE THIS TOOL for repeat-business questions: "what’s our customer retention?", "do buyers come back?", "how do this quarter’s new customers compare to last quarter’s?", "what share of subscribers stay active month over month?". REQUIRES customer-level data: a table whose rows carry a persistent person identifier (customer_id, subscriber_id, email…) — commerce and CRM platforms (Shopify, WooCommerce, Stripe, Klaviyo) qualify; AD PLATFORMS DO NOT (their rows are daily totals about campaigns, not people). Use pma_list_available_analyses to check eligibility, and pma_describe_platform to find the identity column. Counting happens inside the database — individual customer ids/emails never leave it; only aggregate counts are analyzed. Returns each cohort’s size and retention percentages by month offset, plus size-weighted averages over cohorts old enough to measure. Note the stated caveat: cohorts start at the analysis window, so customers first seen BEFORE it are excluded — quote that alongside the numbers. "No repeat activity" is itself a meaningful, reportable answer (one-off purchase businesses exist). DO NOT USE for daily totals or trends → pma_get_trend. DO NOT USE on ad-platform tables — it will refuse (no person identity there).

pma_analyze_correlation

ChatGPT
Tests whether two to six metrics MOVE TOGETHER day to day — within one platform or across platforms (ad spend vs store revenue is the classic) — including who tends to move FIRST (lead-lag up to a week). USE THIS TOOL for relationship questions: "does Facebook spend track Shopify sales?", "do sessions follow ad clicks?", "which platform’s spend moves with total revenue?", "when spend jumps, how soon do orders react?". The math is guarded against the two classic traps: it works on day-over-day CHANGES with each series’ ordinary weekday rhythm removed (so two metrics that merely both grow, or both dip on weekends, do NOT read as related), and every tested lag enters a multiple-testing correction (the default threshold held unrelated real-customer pairs at 0% false positives). CRITICAL FRAMING RULE, report it exactly this way: correlation is ASSOCIATION, not causation. Say "moves with" / "tends to precede" — NEVER "drives", "causes", or "because of". "No significant relationship" is itself a meaningful, reportable answer. If series overlap too little (under ~30 shared days) or fail to fetch, the result says so plainly instead of guessing — report that reason as-is. DO NOT USE to explain a change WITHIN one metric → pma_analyze_drivers. DO NOT USE for side-by-side totals → pma_compare_sources.

pma_analyze_drivers

ChatGPT
Explains WHICH segments (campaigns, countries, devices…) drove a metric’s change between the recent period and the one before it — and for rates, WHETHER the change came from segments performing differently (efficiency) or from volume shifting between segments (mix). USE THIS TOOL for "why did it move?" questions: "why did spend jump?", "which campaigns drove the drop?", "did ROAS fall because ads got worse or because budget moved to weaker campaigns?", "what’s behind the conversion-rate decline?". Plain metric (spend, revenue, clicks): pass metric + dimension — returns each segment’s contribution to the total change, ranked, with new and disappeared segments marked. Rate metric (ctr, cvr, roas): ALSO pass rate_numerator + rate_denominator (the rate’s constituent columns from pma_describe_platform, e.g. clicks + impressions for CTR) — returns the exact efficiency-vs-mix split, which is the actual answer to "worse ads or worse mix?". All arithmetic is exact and deterministic (the two effects sum to the total change). If the data cannot support the analysis (one segment only, empty baseline period), the result says so plainly instead of guessing — report that reason as-is. DO NOT USE for single-day outliers → pma_detect_significant_anomalies. DO NOT USE for a plain period totals comparison → pma_get_date_range_comparison.

pma_compare_sources

ChatGPT
Returns a ranked side-by-side comparison of a single metric across 2–10 platforms in one call. USE THIS TOOL when the user is comparing platforms against each other: "Amazon vs Shopify — which had more orders?", "Compare Stripe and PayPal revenue", "Which ad platform has better ROAS?", "Facebook vs Google vs TikTok spend", "which platform brings in the most contacts — HubSpot or Salesforce?". DO NOT USE WHEN: Comparing campaigns within one platform → use pma_query_custom with dimensions=["campaign_name"]. User needs multiple metrics compared → call pma_query_custom separately for each platform and synthesize. Only one platform is mentioned → use pma_get_account_summary or pma_query_custom. COMPLEX QUESTION: "How are all my platforms doing vs each other on revenue?" → call pma_list_data_sources to get all connected platforms with has_data=true, then call this tool with all of them in one request.

pma_create_dataset

ChatGPT
Create a new dataset — a saved multi-source reporting workspace that blends data from multiple platforms into one view for analysis or export. Two modes are supported: (1) From a template — provide a template_id and source_selections to pre-populate the dataset with data tables. (2) From scratch — provide only a name to create a blank dataset you can add data tables to later. Use pma_list_dataset_templates to find available template IDs and their source group requirements. Use pma_list_data_sources to find valid account IDs for source selections. After creating, use pma_add_data_table to pull in connector data, then pma_get_dataset_data to view the blended results.

pma_describe_platform

ChatGPT
Returns everything you need to query a platform: all valid report_type values AND — when you provide a report_type — the full list of queryable metrics and dimensions for that report. CALL THIS WHEN: (1) You need to find valid report_type values before calling pma_query_custom, pma_query_performance, or pma_detect_anomalies. (2) A metric or field returned null and you need to find the correct column name. (3) The user asks what data is available for a platform ("what can I query for Shopify?", "what fields does HubSpot have?"). SKIP THIS CALL only if you already called it earlier in the same session for the same connector_type + report_type. Never skip when you plan to pass explicit metrics or dimensions to pma_query_custom — always verify column names first. FORMULA FIELDS (is_formula: true) are computed from their component fields. pma_query_custom serves them like any metric (e.g. Shopify "num_of_orders", Google Business Profile "total_review_count" and "starRating_number" on report_type="reviews", grouped by "location_id"). pma_get_trend, pma_forecast, pma_detect_seasonality, pma_detect_changepoints, pma_detect_significant_anomalies and pma_analyze_correlation compute SUM-ratio and row-arithmetic formula shapes; a shape they cannot compute returns an error naming the component fields. pma_detect_anomalies and pma_analyze_drivers need component fields, not formulas. JOINED DIMENSIONS (via_join: true) live on a sibling report the connector joins in - request them together with this report's metrics (e.g. google_ads "campaign_name" or "advertisingChannelType" with clicks/cost on report_type="campaigns_insights"; the join happens server-side). Always copy connector_type verbatim from pma_list_data_sources.

pma_detect_anomalies

ChatGPT
Detects statistical anomalies — days where a metric deviated significantly from its rolling baseline (mean ± std dev). USE THIS TOOL when the user asks about unusual patterns, spikes, drops, or unexpected changes: "anything weird/unusual/abnormal with my X?", "any spikes in X?", "sudden drop in X?", "flag anything unexpected", "did anything change recently?", "out of the ordinary", "alert me to anything strange", "why did X go up/down suddenly?". DO NOT use pma_query_custom or pma_query_performance for these questions — this tool is purpose-built for anomaly detection and computes statistical z-scores that a simple query cannot replicate. COMPLEX QUESTION: "Give me a full health report including anything unusual" → run pma_get_account_summary for the overview, then this tool for anomaly detection, then combine. "Is there anything weird across all my platforms?" → call this tool once per connected platform. PREFER pma_detect_significant_anomalies when the user wants trustworthy, seasonality-aware outliers with a confidence attached — this tool's rolling z-score flags ordinary weekly troughs too, and carries no significance test. This tool remains the quick scan (multi-metric, Mahalanobis, zero-gaps). Uses a lookback window to build a baseline — the longer the lookback, the more stable the baseline. sensitivity="high" catches subtle shifts; sensitivity="low" flags only severe outliers. GAPS vs ANOMALIES: Days where the metric was zero are classified as data gaps (data_gaps / gap_count) and are NOT included in anomalies. Always report gaps separately to the user — "X days had no data reported (possible sync gap)" — so they are not mistaken for unusual drops.

pma_detect_changepoints

ChatGPT
Finds sustained level shifts in a metric’s daily history (PELT changepoint detection) — the dates where the metric stepped to a new normal and stayed there. USE THIS TOOL when the user asks when a metric changed level and by how much: "when did my costs go up?", "did spend step to a new level recently?", "my revenue seems permanently lower — since when?", "did the campaign change in March stick?", "has my cost base shifted?". Returns each detected shift with its date, the average before and after, the change amount and percent, plus the stable segments between shifts — all computed deterministically by our analysis engine, never estimated. If the data cannot support the analysis (too little history, too many missing days, not a daily table), the result says so plainly instead of guessing — report that reason to the user as-is. Use pma_list_available_analyses first when you are unsure whether the data qualifies. DO NOT USE for single-day spikes or dips → pma_detect_anomalies. DO NOT USE for recurring weekday/monthly patterns → pma_detect_seasonality.

pma_detect_seasonality

ChatGPT
Separates a metric’s daily history into trend, seasonal pattern, and noise (STL decomposition), so recurring weekday patterns are not mistaken for real changes. USE THIS TOOL when the user asks whether a movement is a real change or just a recurring pattern: "is the Monday dip normal?", "do my sales always drop on weekends?", "is there a weekly pattern in my spend?", "which day of the week performs best?", "is this drop seasonal or a real problem?". Returns seasonal strength and trend strength (0-1), each weekday’s typical % above or below the average, and the peak and trough weekdays — all computed deterministically by our analysis engine, never estimated. If the data cannot support the analysis (too little history, too many missing days, not a daily table), the result says so plainly instead of guessing — report that reason to the user as-is. Use pma_list_available_analyses first when you are unsure whether the data qualifies. DO NOT USE for single-day spikes or outliers → pma_detect_anomalies. DO NOT USE for sustained level shifts ("did spend step up in March?") → pma_detect_changepoints.

pma_detect_significant_anomalies

ChatGPT
Finds days that are STATISTICALLY unusual after the weekly pattern is removed — each flagged day comes with the expected value, how far the actual landed from it, and a confidence, so an ordinary weekend dip never reads as a problem. USE THIS TOOL when the user wants trustworthy outliers: "which days were genuinely unusual?", "was that spike real or just a busy Friday?", "flag anomalies but skip the normal weekly dips", "how confident are we that day was abnormal?". Set metric_family="count" for whole-number metrics (clicks, orders, impressions, sessions) — they get a count-appropriate statistical test instead of the continuous one. Returns each significant day with its value, the expected value, % deviation, and a confidence (capped at 99.9%) — all computed deterministically by our analysis engine. "No significant anomalies" is itself a meaningful, reportable answer. If the data cannot support the analysis (too little history, too many missing days, not a daily table), the result says so plainly instead of guessing — report that reason as-is. Use pma_list_available_analyses first when you are unsure whether the data qualifies. DO NOT USE for a quick unfiltered outlier scan → pma_detect_anomalies (rolling z-score, no seasonal adjustment, no significance). DO NOT USE for sustained level shifts ("did spend step up in March?") → pma_detect_changepoints.

pma_forecast

ChatGPT
Projects a metric’s next days with an uncertainty range, and answers "are we on pace?" for the current month: the projected month total (actuals so far + forecast of the remaining days) compared against last month’s actual. USE THIS TOOL when the user asks about pace or expectations: "are we on track this month?", "is this month’s dip real or is it just early?", "what will spend look like next week?", "project revenue for the rest of the month". This is the RIGHT answer when a partial month looks like a collapse — never quote a partial-month percent drop as a trend when this tool can give the projected total instead. Returns per-day projections with prediction intervals (default 80%, empirically calibrated on real customer series), the horizon total, and the month pacing comparison — all computed deterministically by our analysis engine. ALWAYS present projected figures as projections ("projected", "on pace for"), never as actuals; the uncertainty range widens with the horizon, so quote the interval alongside the point figure. If the data cannot support a forecast (under 4 weeks of history, too many gaps, not a daily table), the result says so plainly instead of guessing — report that reason as-is. Use pma_list_available_analyses first when you are unsure whether the data qualifies. DO NOT USE for what already happened → pma_get_trend. DO NOT USE to explain WHY a metric moved → pma_analyze_drivers.

pma_get_account_summary

ChatGPT
Returns a single aggregated row of key performance totals (spend, conversions, revenue, ROAS, row count) for a platform and date range. USE THIS TOOL when the user wants a high-level snapshot with no breakdown: "how is my X doing", "what's my total X revenue", "summarise my X", "give me an overview of X", "how much did I spend on X last month", "what are my Salesforce totals", "how many Stripe charges". Works for ANY platform — ad platforms (Facebook Ads, Google Ads), e-commerce (Shopify, Amazon, WooCommerce), CRM (HubSpot, Salesforce), payments (Stripe, PayPal), and more. Omit connector_type to get a summary across ALL connected platforms in one call. DO NOT USE WHEN: User needs a breakdown by campaign, product, category, or any dimension → use pma_query_custom. User wants to see individual records → use pma_query_performance. User is comparing platforms head-to-head → use pma_compare_sources. THIS TOOL SERVES EXACTLY THREE SLOTS: spend, conversions, and conversion value (plus ROAS from them). For ANY other additive metric - reach, impressions, engagement, sessions - use pma_query_custom with aggregate_only=true instead; a summary that comes back without your metric is not evidence the data lacks it. A rate like frequency or CTR comes from its formula field or its components (e.g. impressions and reach), never from summing the rate itself. If a metric comes back null, call pma_describe_platform with field_search to find the correct column name (e.g. pma_describe_platform with connector_type="shopify", report_type="orders", field_search="price"). COMMON FIELD MAPPINGS: Ad platforms (Facebook, TikTok, Bing): spend_field="spend", conversions_field="conversions", conversion_value_field="conversion_value". Google Ads: spend_field="cost", conversion_value_field="conversions_value". Amazon Ads: spend_field="cost", conversions_field="attributedConversions14d", conversion_value_field="attributedSales14d" ("sales" and "orders" are formula fields and cannot be summed). Shopify: report_type="orders", conversions_field="orders", conversion_value_field="order_total_amount" (no spend column, so ROAS is not meaningful). Stripe: report_type="transactions", conversion_value_field="transaction_amount" (no spend column, so ROAS is not meaningful). Field names are validated against the platform schema — a name that does not exist is dropped from the request (its value returns null with a note) instead of failing the whole call, and the request is kept to fields that share one report (a field on a different report is dropped with a note naming the report to ask for it on). A report_type the platform does not have is ignored with a note listing the real ones — the fields still fetch.

pma_get_data_table_data

ChatGPT
Retrieve raw data rows from a single data table within a dataset. A data table is a saved query for one connector account — it captures data from one specific platform and account using pre-configured fields and date range. Use this when the user wants to inspect the data for a specific source inside a dataset. Returns all available rows — use pma_get_dataset_data with a limit parameter if the result set is too large for the context window. Get the data_table_id from pma_get_dataset. Use pma_get_dataset_data instead if you want data blended across all data tables in a dataset.

pma_get_dataset

ChatGPT
Get details about a specific dataset including all its data tables. A data table is a saved query for one connector account inside the dataset — it defines which fields and date range to pull from that source. Returns the dataset name, creation date, and the list of data tables with their connectors and field lists. Use this when the user asks about a specific dataset or wants to see what data tables it contains. Get the dataset_id from pma_list_datasets.

pma_get_dataset_data

ChatGPT
Retrieve the blended data rows from a dataset — runs all data tables in the dataset and returns their combined results. Use this when the user wants to see the actual data inside a dataset, or wants to compare data across platforms in one result set. Get the dataset_id from pma_list_datasets. Specify a date range to filter results — use YYYY-MM-DD format. Use pma_get_data_table_data instead if you want data from one specific connector source within the dataset.

pma_get_date_range_comparison

ChatGPT
USE THIS TOOL FIRST — before pma_query_custom — for ANY period-over-period question. DO NOT use pma_query_custom for period comparisons even when an account_id is specified. TRIGGERS — immediately use this tool when you see any of: "vs last week/month/quarter/year", "week over week", "month over month", "month-on-month", "monthly comparison", "WoW", "MoM", "QoQ", "YoY", "how did X change", "is X up or down", "am I spending more or less", "compare this period to last", "how does this month compare", "monthly trend", "year-on-year", "quarterly comparison". Supports optional account_id to scope to a single account — pass it verbatim from pma_list_data_sources. When account_id is omitted, compares across all accounts of that connector type. Returns the current period value, previous period value, and the absolute and percentage change for a single metric. Run multiple calls in parallel to compare several metrics at once. Preset comparison modes (no date math needed): "wow" = last 7 days vs prior 7 days. "mom" = last 30 days vs prior 30 days. "qoq" = last 90 days vs prior 90 days. "yoy" = last 365 days vs prior 365 days. "custom" = supply all four dates manually. DO NOT USE WHEN: Comparing two different platforms → pma_compare_sources. User wants a breakdown by campaign or dimension WITHIN one period → pma_query_custom. User wants raw row data → pma_query_performance. If the result comes back null, call pma_describe_platform to find the correct metric name (e.g. Shopify revenue may be "total_price" not "spend").

pma_get_organization_info

ChatGPT
Get identity and status information for the connected hub. Returns the organization_id (unique identifier), hub_name (human-readable name), subscription plan status, and total_accounts count. Use this when the user asks "what hub am I connected to?", "what is my hub name?", "what is my organization ID?", "which organization is this?", or any question about hub identity. For the full account list, use pma_list_data_sources.

pma_get_token_health_summary

ChatGPT
Returns an org-wide snapshot of provider OAuth token health across all connected accounts. Use this when the user asks "why is my data not syncing?", "are all my connections healthy?", or "which accounts have issues?" — before diving into per-account details. Returns: summary counts (total, valid, expired, revoked, expiring soon), at_risk_accounts list with specific action_needed guidance per account, and a single recommendation sentence to relay to the user. For each at-risk account, call pma_list_usages to see which sub-accounts are affected. For the full account list with pagination, call pma_list_accounts_with_token_status.

pma_get_trend

ChatGPT
USE THIS TOOL when the user wants to see how a metric moved over multiple periods — days, weeks, or months — within a date range. TRIGGERS: "show me monthly spend over the last 6 months", "how has my ROAS trended over the year", "weekly clicks for Q1", "daily conversions this month", "month-by-month breakdown of X", "trend of X over time". Returns one row per period (day/week/month) with the metric value AND pre-computed change vs the prior period — no manual arithmetic needed. Supports optional account_id to scope to a single account. Run multiple calls in parallel to trend several metrics at once. DO NOT USE WHEN: User wants to compare exactly two periods ("MoM", "vs last month") → pma_get_date_range_comparison. User wants a breakdown by campaign or dimension within one period → pma_query_custom. User wants raw individual rows → pma_query_performance.

pma_inspect_org_data

ChatGPT
Diagnostic tool — scans connected platforms and report types to show exactly what data is available. For each connector × report type combination, returns: whether data exists, total row count, earliest and latest dates in the table, and how many accounts have synced data. USE THIS TOOL WHEN: A query returns no results and you need to confirm whether data has been synced at all. The user asks "do I have data from X?", "is my X connected and working?", "why is X not returning results?". You need to find which report types have data before deciding what to query. Debugging a connector that appears connected but is returning nothing. PERFORMANCE: Pass connector_type to inspect a single platform (fast, ~5s). Omit connector_type to scan ALL platforms (slow, may take 30-45s). DO NOT call the full scan routinely before every query — use pma_list_data_sources (has_data flag) instead.

pma_list_accounts_with_token_status

ChatGPT
Returns all connected accounts for this organization with their provider OAuth token health. Use this when the user asks why data is not syncing, which accounts have connection issues, or to find a specific account by name before calling pma_list_usages or pma_query_account. PAGINATION — CRITICAL: This tool returns accounts in pages (default 100 per page). Organizations may have many connected accounts across platforms. If the account you are looking for is NOT found on the first page AND has_more is true, you MUST call again with offset += limit to get the next page. NEVER tell the user an account does not exist until you have checked ALL pages (has_more: false). token_status values: "valid" = token is active and data sync will work. "expired_needs_refresh" = token has expired; data sync is paused until the account is re-connected. "invalid_revoked" = credentials were permanently revoked by the provider; the user must re-connect the account in the PMA dashboard. Each account also shows usages_count (number of sub-accounts e.g. ad accounts, GA4 properties). Call pma_list_usages for a specific account to see per-sub-account access status. Call pma_get_token_health_summary for an at-a-glance org-wide health check. RESPONSE FORMAT — when the user asks to list their accounts or check connection health, ALWAYS use this exact structure (plain text, no markdown tables, no emojis): Connected accounts: N total (N healthy, N need attention) Healthy (N): - Platform Name — Account Name (email@example.com) Need attention (N): - Platform Name — Account Name [token expired — reconnect in PMA to resume syncing] - Platform Name — Account Name [access revoked — reconnect in PMA to resume syncing] Rules: summary line first; healthy accounts first, then "need attention"; omit a section entirely if its count is 0; translate token_status — "valid" → healthy, "expired_needs_refresh" → [token expired — reconnect in PMA], "invalid_revoked" → [access revoked — reconnect in PMA]; never show raw token_status codes; if has_more is true note "Showing first N of M accounts."

pma_list_activity

ChatGPT
List recent activity events for this organization with pagination. Returns events like data syncs, exports, and connector changes. Use this when the user asks about recent activity, what happened recently, or wants to review organization events. Results are paginated — use the cursor from a previous response to get the next page.

pma_list_available_analyses

ChatGPT
Checks which statistical analyses (seasonality, level-shift detection, and any our analysis engine adds in the future) can run on a specific platform table and metric — and explains, in plain language, why the rest cannot. USE THIS TOOL before pma_detect_seasonality / pma_detect_changepoints when you are unsure the data qualifies, or when the user asks what is possible: "what analyses can you run on my Shopify orders?", "can you check seasonality on this?", "why can’t you decompose this metric?". The check is three-layered: first the table’s declared kind (snapshot tables — lifetime or point-in-time captures — never qualify for time-series analyses), then the actual data’s shape (how many days of history, how many gaps), each analysis judged against its own declared requirements. Anything the check cannot verify is listed honestly as unverified. Returns, per analysis: eligible true/false, the reasons when not, and the requirements it was judged against. Quote the reasons to the user as-is — they are written for that. This tool only CHECKS eligibility — run the analysis itself with the matching pma_detect_* tool. If the analysis engine is not deployed yet, the result says so; that means no engine-backed analyses are available anywhere, not that this data is unsuitable.

pma_list_data_sources

ChatGPT
ALWAYS call this first — before any other tool — whenever you do not already know what platforms are connected or what their exact connector_type values are. Returns every connected account grouped by platform, with: connector_type (exact string to copy into all other tools), account_id (exact string for pma_query_custom with account_id), display name, email, connected_at date, last_sync date (when the sync job last ran), data_through date (the latest date the warehouse actually has data for — may be yesterday even if last_sync is today, because syncs collect data through the previous day), and has_data flag (true = data exists and is queryable). CRITICAL — interpret has_data and last_sync carefully: has_data: true means Firestore sync records exist — safe to query. has_data: false means no Firestore sync records, but SQL data may still exist (sync tracking can be cleared or missing). If has_data is false and the account has been connected for a while, still try querying — use pma_inspect_org_data to verify if unsure. last_sync shows the most recent sync date — if it is months ago, widen the date range in your query beyond the default 30 days. PAGINATION WARNING — a platform like Facebook is stored as multiple connector types (facebook, facebook_ads, facebook_instagram, facebook_pages) that may appear on DIFFERENT pages. NEVER conclude a platform has no data after reading only page 1. When the user asks about a specific platform (e.g. "Facebook", "Google"), ALWAYS pass connector_type_filter to get ALL variants in one call. When doing a general listing and has_more is true, paginate ALL pages before drawing conclusions. NAMED CLIENT LOOKUP - when the user names a specific client, store, brand or account, pass name_filter to resolve that name across EVERY platform in one call: it searches account names, emails AND sub-account display names, and returns the matching ids plus the column to filter queries by. COMPLEX QUESTION PROTOCOL — when a user asks a broad or multi-platform question: (1) Call this tool first to get all connected platforms. (2) Filter to platforms where has_data is true. (3) Route each platform to the right tool: totals → pma_get_account_summary, analysis/grouping → pma_query_custom, comparison → pma_compare_sources, anomalies → pma_detect_anomalies. (4) Synthesize results across platforms into one coherent answer. RESPONSE FORMAT — when the user asks to list or see their sources/platforms, ALWAYS use this exact structure (plain text, no markdown tables, no emojis): Connected sources: N platforms, N accounts Platform Name (N account(s)) - Account Name (email@example.com) — data through: YYYY-MM-DD — ready - Account Name — data through: YYYY-MM-DD — no data yet Rules: summary line first; one group per platform; bullet per account with name, email if present, data_through date (if non-null, prefer over last_sync for display; if null, fall back to last_sync), then "ready" or "no data yet"; never show internal connector_type codes; if has_more is true note "Showing first N of M platforms. Ask me to continue for the rest."

pma_list_dataset_templates

ChatGPT
List pre-built dataset recipes for common marketing stacks (e.g. "Facebook Ads + Google Ads performance overview"). Templates skip manual field selection and data table setup — the dataset is pre-configured with the right connectors and fields for the chosen marketing use case. ALWAYS use this tool when the user wants to: create a dataset, convert data into a dataset, build a dataset from a template, or choose a template for dataset creation. Also use this when the user asks how to combine data from multiple platforms or wants a ready-made reporting setup. Returns template IDs, names, descriptions, platforms, categories, source group requirements, and data table count. The id field from each result is what you pass as template_id to pma_create_dataset. Source groups show which connector types the user must have connected to use the template. A source group with maxAccounts accepts at most that many accounts; ask the user which ones to use rather than selecting all.

pma_list_datasets

ChatGPT
A dataset is a saved multi-source reporting workspace — it groups data from multiple platforms (e.g. Facebook Ads + Google Ads) into one blended view for analysis or export to Google Sheets, Data Studio, or Power BI. Lists all datasets for this organisation. Returns dataset names, IDs, and a summary of the data tables inside each dataset. Use this first to discover available dataset IDs before calling pma_get_dataset or pma_get_dataset_data. Use this when the user asks "what datasets do I have", "show me my datasets", or wants to work with a specific dataset. Suggest creating a dataset when the user wants to compare data across multiple platforms, build a report they will run repeatedly, or export blended data to Google Sheets or Data Studio.

pma_list_usages

ChatGPT
Returns the individual data sources (called "usages" or "sub-accounts") that sit under one parent connected account — for example, the specific Facebook ad accounts, Google Ads customer IDs, GA4 properties, or Bing advertiser accounts linked to a single login. WHEN TO USE: (1) The user asks "which ad accounts / properties / stores does [account name] have access to?" (2) You need to find a specific sub-account name or usage_id before querying data for it with pma_query_account. (3) You are diagnosing why a specific sub-account has no data — check access_status and reports[].status. (4) After pma_list_accounts_with_token_status shows usages_count > 0 and you need the individual sub-account details. WHEN NOT TO USE: Do NOT call this before every query — you only need it when you specifically want the sub-account list. If you already have an account_id and just want data, go directly to pma_query_custom (pass account_id). If you want org-wide token health, use pma_get_token_health_summary instead. If you want to know which platforms are connected (not the sub-accounts within them), use pma_list_data_sources. WHAT IT RETURNS: Each usage entry has: usage_id (the sub-account ID — use this in pma_query_custom with account_id to scope queries to one sub-account), name (display name e.g. "Greek Market", "UK Campaigns"), is_active (whether this sub-account is currently syncing), access_status ("accessible" = provider grants access; "lost" = user revoked PMA permissions at sub-account level), last_sync (most recent sync date, YYYY-MM-DD), reports (per-report-type sync status: "success", "syncing", or "error" with last_successful_sync date). The response also includes parent_account.token_status so you can distinguish a sub-account access problem from a parent token problem in a single call. PREREQUISITES: Call pma_list_data_sources first to get the connector_type (e.g. "facebook_ads") and account_id (e.g. "facebook-1516721155103340") — both must be copied verbatim from that response. If the user only knows the email address of the account owner, call pma_list_accounts_with_token_status with email_filter first to find the matching account_id, then pass that account_id here.

pma_query_custom

ChatGPT
Use for analysis, grouping, ranking, or trends WITHIN a single time window on a single platform. USE THIS TOOL for: "top N [campaigns/products/deals] by [metric]" → dimensions=["name"], order_by="metric". "which ACCOUNT/store spent or earned the most" → dimensions=["account_name"], order_by="metric" - ONE grouped call ranks every connected account (works on every platform; rows carry account names), NEVER one call per account. "breakdown of X grouped by month/week/day within one period" → dimensions=["date"], aggregation="sum". "breakdown of X by [category/status/channel]" → dimensions=["category"]. "how many / how much X" → metrics=["count"/"revenue"], no dimensions for a grand total. "which [item] has the highest/lowest X" → order_by="metric", limit=1. "trend of X over time" → dimensions=["date"], order_direction="asc". COMPLEX MULTI-PART QUESTIONS — run multiple calls in sequence or parallel: "Give me a full marketing report" → (1) pma_get_account_summary for each platform overview, (2) pma_query_custom for top campaigns per platform, (3) pma_compare_sources for cross-platform totals, (4) pma_detect_anomalies if the user also wants to know about unusual activity. "Show me my best campaigns and flag anything unusual" → pma_query_custom (top campaigns) + pma_detect_anomalies (anomalies). SCHEMA AUTO-DISCOVERY: automatically fetches available columns before running — if metrics is omitted, up to 5 are auto-selected. DO NOT USE WHEN: User asks about change between two periods ("MoM", "WoW", "vs last month", "month-on-month", "month over month", "how has X changed", "is X up or down", "am I spending more this month") → pma_get_date_range_comparison. User wants to SEE individual records with no aggregation → pma_query_performance. User names a specific account → this tool with account_id (metrics/dimensions for its aggregates, fields[] for its raw records). User is comparing 2+ platforms on the same metric → pma_compare_sources. User asks about spikes, drops, or anything unusual → pma_detect_anomalies. TIMEOUT RECOVERY: If a query times out (common for Facebook Ads same-day data), use pma_create_dataset to save these query parameters as a dataset. The user can then access the data through Google Sheets or Data Studio, which can handle longer API response times.

pma_query_performance

ChatGPT
Returns raw individual data rows from a platform — no aggregation, no totals, exactly as stored. For platforms with multiple subtypes (e.g. Facebook → facebook_ads / facebook_instagram / facebook_pages, or Google → google_ads / google_analytics), this tool may return 0 rows due to connector ambiguity — use pma_query_custom with a specific account_id instead, which always resolves correctly. If this tool returns 0 rows, call pma_list_data_sources to confirm accounts are connected and have has_data: true. USE THIS TOOL ONLY when the user explicitly wants to browse or inspect records: "show me", "list", "display", "pull up", "give me the last N", "what are my recent X". Examples: "show me the last 20 Shopify orders", "list my active Facebook campaigns", "pull up recent Stripe transactions", "display my latest HubSpot deals". HARD STOP — DO NOT USE when the answer involves any of these — use pma_query_custom instead: "how many", "how much", "total", "sum", "average", "top N by [metric]", "grouped by", "per month/week/day", "breakdown", "which has the most/least", "compare", "trend". If the user names a SPECIFIC account by name or ID → use pma_query_custom with account_id instead (this tool queries ALL accounts together). By default returns the first 5 columns only. Specify fields to get particular columns.

Capabilities

Writes

Example Prompts

Click any prompt to copy it.

Similar Apps

Listed on MCP App Store

Listed on MCP App Store

Add this badge to your site or README. It links to this page.

HTML

<a href="https://mcp-app-store.keenethics-labs.com/app/power-my-analytics"><img src="https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg" alt="Listed on MCP App Store" width="220" height="44"></a>

Markdown

[![Listed on MCP App Store](https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg)](https://mcp-app-store.keenethics-labs.com/app/power-my-analytics)