Google BigQuery

@Google BigQuery run a query to show the top 5 products by revenue from our sales table this quarter

The query returned five rows from your `sales.products_summary` table. The top product generated $847,320 in Q1 revenue, followed by four others ranging from $612,450 to $498,100. Each row includes product_id, product_name, total_revenue, and units_sold. This example assumes your BigQuery project contains a table with those columns and recent transaction data.

This read-heavy example demonstrates basic SELECT queries with aggregation and filtering. You'll need OAuth scopes that grant read access to the specific BigQuery datasets and tables you reference. Query costs depend on the amount of data scanned—preview your table schema to estimate bytes processed before running large queries.

@Google BigQuery create a new table called `user_cohorts` in the analytics dataset with columns user_id, cohort_month, and first_purchase_date

A CREATE TABLE statement was executed successfully in your `analytics` dataset. The new `user_cohorts` table now exists with three columns: user_id (STRING), cohort_month (DATE), and first_purchase_date (TIMESTAMP). This example illustrates a write operation—no rows were inserted yet, but the schema is ready for subsequent INSERT or MERGE queries.

This write example shows DDL (Data Definition Language) capability. You must have BigQuery Data Editor or Owner permissions on the target dataset. Be cautious: CREATE TABLE will fail if the table already exists unless you add IF NOT EXISTS. Consider using CREATE OR REPLACE TABLE if you want to overwrite existing schemas during iterative development.

@Google BigQuery analyze our event logs from the past week and summarize which feature flags correlate with the highest session duration

The query joined your `events.sessions` and `events.feature_flags` tables, filtering to the last seven days. Results show that users with the 'new_checkout_flow' flag enabled averaged 8.2 minutes per session, compared to 5.1 minutes for the control group. Two other flags ('dark_mode', 'recommendations_v2') showed smaller positive correlations. This synthesis pairs BigQuery's aggregation with AI interpretation to surface actionable insights from raw event data.

This synthesis example demonstrates how the AI can construct multi-table JOIN queries and interpret statistical patterns. Accuracy depends on your schema design—ensure foreign keys and timestamps are indexed for performance. Large event tables may incur significant query costs; consider partitioning by date and clustering by user_id to reduce bytes scanned and speed up analysis.

When BigQuery MCP beats exporting CSVs for finance teams

A 6-person finance team runs weekly revenue reviews where the CFO asks follow-up questions that weren't in the pre-built dashboard. Instead of waiting for the data analyst to export new CSVs, the team uses Switchy with BigQuery MCP to query transaction tables directly during the meeting. The OAuth2 flow means everyone authenticates once with their Google Workspace account and inherits the same row-level security their org already configured. This works when your queries are straightforward aggregations or filters—if you need complex joins across 8 tables or window functions, you're better off pre-building those in dbt and querying the mart. The single-line query constraint is annoying for readability but doesn't block real work. If your exec reviews involve live data questions and your source of truth is already in BigQuery, this MCP closes the loop without leaving the conversation.

Why BigQuery MCP works for support teams with data in warehouses

A 12-person support team handles tier-2 escalations where the answer isn't in Zendesk—it's buried in event logs or subscription tables that live in BigQuery. Support leads use Switchy to query customer activity without pinging engineering every time. The OAuth2 setup means each support agent authenticates with their own Google account, so access control stays consistent with what IT already provisioned. This scenario breaks down if your queries need more than one tool—BigQuery MCP only exposes a single query tool, so you can't list tables, inspect schemas, or preview results separately. If your support team runs the same 5 queries over and over, build a Retool app instead. But for one-off escalation lookups where the question changes each time, BigQuery MCP keeps support moving without opening the GCP console or filing a Slack request to data eng.

When BigQuery MCP is overkill for small marketing teams

A 3-person marketing team wants to understand which channels drove signups last quarter. They have Google Analytics 4 data exported to BigQuery and think the MCP will let them ask natural-language questions during their retro. The problem: with only one query tool and no schema introspection, they spend more time debugging SQL syntax errors than analyzing results. The single-line format makes it hard to read queries with multiple joins, and OAuth2 setup is friction when the team doesn't already live in Google Workspace. If your marketing team isn't SQL-fluent or your BigQuery schema changes often, you're better off using Looker Studio or a BI tool with a semantic layer. BigQuery MCP makes sense when your team already writes SQL daily and needs ad-hoc access during live conversations—not when you're learning SQL for the first time or running the same reports every week.