What is BigQuery MCP Server?
BigQuery MCP Server is a Model Context Protocol (MCP) server designed to facilitate access to Google BigQuery, enabling Large Language Models (LLMs) to comprehend BigQuery dataset structures and execute SQL queries.
How to use BigQuery MCP Server?
To use the server, you can either install it locally or run it in a Docker container. After installation, configure it with your Google Cloud project ID and authentication credentials, then use the provided tools to execute queries or list datasets and tables.
Key features of BigQuery MCP Server?
- Supports Application Default Credentials and service account key files for authentication.
- Tools for executing SQL queries, listing datasets, and retrieving table information.
- Security features to ensure only read-only access and prevent excessive costs.
Use cases of BigQuery MCP Server?
- Executing SQL queries to retrieve data from BigQuery.
- Listing all datasets and tables for data exploration.
- Validating queries without execution to estimate costs.
FAQ from BigQuery MCP Server?
- What types of queries can I execute?
Only read-only (SELECT) queries are allowed to ensure data security.
- How do I authenticate?
You can authenticate using Application Default Credentials or a service account key file.
- Is there a limit on query processing?
Yes, there is a default limit of 500GB for query processing to prevent excessive costs.
BigQuery MCP Server
A Model Context Protocol (MCP) server for accessing Google BigQuery. This server enables Large Language Models (LLMs) to understand BigQuery dataset structures and execute SQL queries.
Features
Authentication and Connection Management
- Supports Application Default Credentials (ADC) or service account key files
- Configurable project ID and location settings
- Authentication verification on startup
Tools
-
query
- Execute read-only (SELECT) BigQuery SQL queries
- Configurable maximum results and bytes billed
- Security checks to prevent non-SELECT queries
-
list_all_datasets
- List all datasets in the project
- Returns an array of dataset IDs
-
list_all_tables_with_dataset
- List all tables in a specific dataset with their schemas
- Requires a datasetId parameter
- Returns table IDs, schemas, time partitioning information, and descriptions
-
get_table_information
- Get table schema and sample data (up to 20 rows)
- Support for partitioned tables with partition filters
- Warnings for queries on partitioned tables without filters
-
dry_run_query
- Check query validity and estimate cost without execution
- Returns processing size and estimated cost
Security Features
- Only SELECT queries are allowed (read-only access)
- Default limit of 500GB for query processing to prevent excessive costs
- Partition filter recommendations for partitioned tables
- Secure handling of authentication credentials
Installation
Local Installation
# Clone the repository
git clone https://github.com/yourusername/bigquery-mcp-server.git
cd bigquery-mcp-server
# Install dependencies
bun install
# Build the server
bun run build
# Install command to your own path.
cp dist/bigquery-mcp-server /path/to/your_place
Docker Installation
You can also run the server in a Docker container:
# Build the Docker image
docker build -t bigquery-mcp-server .
# Run the container
docker run -it --rm \
bigquery-mcp-server \
--project-id=your-project-id
Or using Docker Compose:
# Edit docker-compose.yml to set your project ID and other options
# Then run:
docker-compose up
MCP Configuration
To use this server with an MCP-enabled LLM, add it to your MCP configuration:
{
"mcpServers": {
"BigQuery": {
"command": "/path/to/dist/bigquery-mcp-server",
"args": [
"--project-id",
"your-project-id",
"--location",
"asia-northeast1",
"--max-results",
"1000",
"--max-bytes-billed",
"500000000000"
],
"env": {
"GOOGLE_APPLICATION_CREDENTIALS": "/path/to/service-account-key.json"
}
}
}
}
You can also use Application Default Credentials instead of a service account key file:
{
"mcpServers": {
"BigQuery": {
"command": "/path/to/dist/bigquery-mcp-server",
"args": [
"--project-id",
"your-project-id",
"--location",
"asia-northeast1",
"--max-results",
"1000",
"--max-bytes-billed",
"500000000000"
]
}
}
}
Setting up Application Default Credentials
To authenticate using Application Default Credentials:
-
Install the Google Cloud SDK if you haven't already:
# For macOS brew install --cask google-cloud-sdk # For other platforms, see: https://cloud.google.com/sdk/docs/install
-
Run the authentication command:
gcloud auth application-default login
-
Follow the prompts to log in with your Google account that has access to the BigQuery project.
-
The credentials will be saved to your local machine and automatically used by the BigQuery MCP server.
Testing
You can use inspector for testing and debugging.
npx @modelcontextprotocol/inspector dist/bigquery-mcp-server --project-id={{your_own_project}}
Usage
Using the Helper Script
The included run-server.sh
script makes it easy to start the server with common configurations:
# Make the script executable
chmod +x run-server.sh
# Run with Application Default Credentials
./run-server.sh --project-id=your-project-id
# Run with a service account key file
./run-server.sh \
--project-id=your-project-id \
--location=asia-northeast1 \
--key-file=/path/to/service-account-key.json \
--max-results=1000 \
--max-bytes-billed=500000000000
Manual Execution
You can also run the compiled binary directly:
# Run with Application Default Credentials
./dist/bigquery-mcp-server --project-id=your-project-id
# Run with a service account key file
./dist/bigquery-mcp-server \
--project-id=your-project-id \
--location=asia-northeast1 \
--key-file=/path/to/service-account-key.json \
--max-results=1000 \
--max-bytes-billed=500000000000
Example Client
An example Node.js client is included in the examples
directory:
# Make the example executable
chmod +x examples/sample-query.js
# Edit the example to set your project ID
# Then run it
cd examples
./sample-query.js
Command Line Options
--project-id
: Google Cloud project ID (required)--location
: BigQuery location (default: asia-northeast1)--key-file
: Path to service account key file (optional)--max-results
: Maximum rows to return (default: 1000)--max-bytes-billed
: Maximum bytes to process (default: 500000000000, 500GB)
Required Permissions
The service account or user credentials should have one of the following:
roles/bigquery.user
(recommended)
Or both of these:
roles/bigquery.dataViewer
(for reading table data)roles/bigquery.jobUser
(for executing queries)
Example Usage
Query Tool
{
"query": "SELECT * FROM `project.dataset.table` LIMIT 10",
"maxResults": 100
}
List All Datasets Tool
// No parameters required
List All Tables With Dataset Tool
{
"datasetId": "your_dataset"
}
Get Table Information Tool
{
"datasetId": "your_dataset",
"tableId": "your_table",
"partition": "20250101"
}
Dry Run Query Tool
{
"query": "SELECT * FROM `project.dataset.table` WHERE date = '2025-01-01'"
}
Error Handling
The server provides detailed error messages for:
- Authentication failures
- Permission issues
- Invalid queries
- Missing partition filters
- Excessive data processing requests
Code Structure
The server is organized into the following structure:
src/
├── index.ts # Entry point
├── server.ts # BigQueryMcpServer class
├── types.ts # Type definitions
├── tools/ # Tool implementations
│ ├── query.ts # query tool
│ ├── list-datasets.ts # list_all_datasets tool
│ ├── list-tables.ts # list_all_tables_with_dataset tool
│ ├── table-info.ts # get_table_information tool
│ └── dry-run.ts # dry_run_query tool
└── utils/ # Utility functions
├── args-parser.ts # Command line argument parser
└── query-utils.ts # Query validation and response formatting
License
MIT