Google BigQuery MCP server ​

The Google BigQuery MCP server enables LLMs to interact with Google BigQuery for analytics and data warehouse operations through natural conversation. It provides tools to discover datasets and tables, estimate query costs before execution, run parameterized SQL queries, execute data modifications, stream rows, and monitor jobs without requiring direct interaction with the Google BigQuery console.

Uses ​

Use the Google BigQuery MCP server to perform the following actions:

• Discover datasets, tables, and their schemas within a BigQuery project
• Preview table rows at zero query cost to sample data
• Estimate query costs via dry run before executing expensive scans
• Run parameterized read-only GoogleSQL SELECT queries with spend caps
• Retrieve complete result sets by polling long-running queries and paginating large results
• Execute data-modifying SQL statements
• Create, alter, and drop datasets, tables, and views via DDL
• Stream JSON rows into tables with per-row error reporting
• Monitor query and DML job progress and retrieve execution statistics
• Cancel running jobs on a best-effort basis

Example prompts ​

Use the following example prompts to invoke Google BigQuery MCP server tools:

• What datasets are available in the analytics project?
• Show me the schema and row count for the customers table in my analytics dataset.
• Preview the first 10 rows of the events table to see what data is available.
• How many bytes would scanning all columns from the large_transactions table cost?
• Run a query to find the top 10 customers by revenue from the last quarter.
• What are the results of the query with job ID xyz123 in the US region?
• Update all records in the test_table where status equals pending to status equals completed.
• Stream 100 new event rows into the events table and report any failures.
• What is the status of my long-running data load job?
• Cancel the query job that's been running for too long.

Google BigQuery MCP server tools ​

The Google BigQuery MCP server provides the following tools:

Install the Google BigQuery MCP server ​

Complete the following steps to install a prebuilt MCP server to your project:

Google BigQuery connection setup ​

How to use Google BigQuery MCP server tools ​

Refer to the following sections for detailed information on available tools:

list_datasets tool ​

The list_datasets tool lists datasets in a BigQuery project. Your LLM uses this tool to discover what datasets are available in a project, optionally filtered by labels, and to understand the warehouse structure before querying.

Try asking:

• What datasets are available in my project?
• Show me all datasets with the environment label set to production.
• List all datasets in alphabetical order with their descriptions.

get_dataset tool ​

The get_dataset tool retrieves metadata for a specific dataset including location, description, and default expiration settings. Your LLM uses this tool to understand dataset configuration, determine the region for job routing, and verify dataset accessibility before running queries.

Try asking:

• Where is the analytics dataset located?
• What's the description and default expiration for the raw_data dataset?
• Get the metadata for the events dataset to confirm its region.

list_tables tool ​

The list_tables tool lists tables, views, and external tables in a dataset. Your LLM uses this tool to see what tables are available within a dataset and understand what data is accessible for analysis.

Try asking:

• What tables are in the analytics dataset?
• Show me all views in the staging dataset.
• List all external tables available in the data_imports dataset.

get_table tool ​

The get_table tool retrieves a table's full schema, row and byte counts, partitioning and clustering configuration, and streaming-buffer status. Your LLM uses this tool to ground itself in table structure before writing SQL, understand what columns are available, and learn about indexing that can reduce scan costs.

Try asking:

• Show me the schema for the customers table.
• What are the partitioning and clustering columns for the events table?
• How many rows and bytes does the transactions table contain?

preview_table_rows tool ​

The preview_table_rows tool reads sample rows directly from a table without running a query or incurring query charges. Your LLM uses this tool to see actual data values, verify data quality, and sample a table's content without cost or SQL execution overhead.

Try asking:

• Preview the first 20 rows of the users table.
• Show me some examples of data in the customer_events table.
• Get a sample of rows from the products table, showing just the name and price columns.

estimate_query_cost tool ​

The estimate_query_cost tool validates a SQL statement and estimates bytes that would be scanned without running the query, because dry runs don't incur cost. Your LLM uses this tool to check SQL syntax, understand what tables a query touches, and surface cost estimates to the user before expensive scans execute.

Try asking:

• What would this query cost: SELECT * FROM huge_table WHERE year = 2024?
• Estimate the cost of a query joining customers and orders on the last month's data.
• Check if my query is syntactically valid and how many bytes it would scan.

run_query tool ​

The run_query tool runs a read-only GoogleSQL SELECT query. If the query completes quickly, it returns rows immediately. Otherwise it returns a job reference for polling, which the LLM should treat as normal progress rather than an error. Your LLM uses this tool to execute parameterized analytical queries and retrieve results.

Try asking:

• Query the customers table to find all customers created in the last 30 days.
• Run a SELECT statement with a parameter to find orders where the amount exceeds @min_amount.
• Execute a query that joins three tables and returns the top 100 results by revenue.

get_query_results tool ​

The get_query_results tool polls a query job and retrieves paginated result rows by job id and location. The tool reports unfinished jobs as still running rather than as an error. Your LLM uses this tool to fetch results from long-running queries and handle multi-page result sets.

Try asking:

• Get the results of job ID abc123 in the us-central1 region.
• Fetch the next page of results from my query using this page token.
• Poll the status of a query that was still running earlier.

execute_sql tool ​

The execute_sql tool executes a single allowlisted data or schema-modifying SQL statement. Your LLM uses this tool to insert, update, or delete rows, and to create, alter, or drop tables and datasets. The tool requires explicit user confirmation for destructive statements, such as DROP, TRUNCATE, or DELETE, before it executes. Otherwise it runs immediately.

Try asking:

• Insert a new row into the customers table with these values.
• Update all records in the staging table where status equals pending to completed.
• Delete all test data from the events table that were created in the last hour.

insert_rows tool ​

The insert_rows tool streams JSON rows into a BigQuery table with optional duplicate suppression. Your LLM uses this tool to append event or application rows generated in the conversation to a table. The tool reports each row's outcome individually rather than treating the whole batch as a single success or failure.

Try asking:

• Stream these 50 event records into the events table.
• Append new user signup records to the users table and report any failures.
• Insert rows with deduplication IDs to avoid inserting duplicates.

get_job tool ​

The get_job tool retrieves the status, statistics, and errors of a BigQuery job. Your LLM uses this tool to check whether a job is still running and review bytes processed and billed. The tool reports a failure for jobs that encountered errors.

Try asking:

• What's the status of job ID xyz789 in the us region?
• Show me how many bytes my query processed and whether results came from cache.
• Get the error details for a job that failed.

cancel_job tool ​

The cancel_job tool requests best-effort cancellation of a running BigQuery job. Your LLM uses this tool to stop a runaway query before it incurs further costs, and to confirm the final state of a cancelled job.

Try asking:

• Cancel the query job that's been running too long.
• Stop the data load job in the us-central1 region.
• Halt this insert operation before it processes more data.

Getting started ​

View and manage your MCP server tools in the Overview page Tools section. Tool management provides the following capabilities:

• Start tools
• Remove tools
• Edit tools
• Add tools
• Edit your LLM-facing MCP server description