BigQuery MCP Server

This is a Python-based MCP (Model Context Protocol) server that retrieves dataset, table, and schema information from Google Cloud BigQuery, caches it locally, and serves it via MCP. Its primary purpose is to enable generative AI systems to quickly understand BigQuery's structure and execute queries securely.

Key Features

• Metadata Management: Retrieves and caches information about BigQuery datasets, tables, and columns
• Keyword Search: Supports keyword search of cached metadata
• Secure Query Execution: Provides SQL execution capabilities with automatic LIMIT clause insertion and cost control
• File Export: Execute queries and save results to local files in CSV or JSONL format
• MCP Compliance: Offers tools via the Model Context Protocol

MCP Server Tools

Available tools:

1. get_datasets - Retrieves a list of all datasets
2. get_tables - Retrieves all tables within a specified dataset (requires dataset_id, optionally accepts project_id)
3. search_metadata - Searches metadata for datasets, tables, and columns
4. execute_query - Safely executes BigQuery SQL queries with automatic LIMIT clause insertion and cost control
5. check_query_scan_amount - Retrieves the scan amount for BigQuery SQL queries
6. save_query_result - Executes BigQuery SQL queries and saves results to local files (CSV or JSONL format)

Tool Details

save_query_result

The save_query_result tool provides advanced query execution with file export capabilities:

Parameters:

• sql (required): SQL query to execute
• output_path (required): Local file path to save results
• format (optional): Output format - "csv" (default) or "jsonl"
• project_id (optional): Target GCP project ID
• include_header (optional): Include header row in CSV output (default: true)

Key Features:

• No Automatic LIMIT: Unlike execute_query, this tool does not automatically add LIMIT clauses to your SQL queries
• Cost Control: Maintains scan amount limits (default: 1GB) and safety checks to prevent expensive queries
• Security: Path validation prevents directory traversal attacks
• Flexible Formats: Supports both CSV and JSONL output formats
• Large Dataset Support: Handles large query results efficiently within scan limits

Example Usage:

-- Export all rows without LIMIT restriction (subject to scan amount limits)
SELECT customer_id, order_date, total_amount 
FROM `project.dataset.orders` 
WHERE order_date >= '2024-01-01'

Important Note: While this tool doesn't add LIMIT clauses, it still enforces scan amount limits for cost protection. Queries that would scan more than the configured limit (default: 1GB) will be rejected.

Installation and Environment Setup

Prerequisites

• Python 3.11 or later
• Google Cloud Platform account
• GCP project with BigQuery API enabled

Install

uv

uv add bq_mcp_server

pip

pip install bq_mcp_server

Installing Dependencies

This project uses uv for package management:

# Install uv if not already installed
curl -LsSf https://astral.sh/uv/install.sh | sh

# Install dependencies
uv sync

Configuring Option

For a list of configuration values, see:

docs/settings.md

MCP Setting

Claude Code

claude mcp add bq_mcp_server -- uvx --from git+https://github.com/takada-at/bq_mcp_server bq_mcp_server --project-ids <your project ids>

JSON

{
    "mcpServers": {
        "bq_mcp_server": {
            "command": "uvx",
            "args": [
                "--from",
                "git+https://github.com/takada-at/bq_mcp_server",
                "bq_mcp_server",
                "--project-ids",
                "<your project ids>"
            ]
        }
    }
}

Running Tests

Running All Tests

pytest

Running Specific Test Files

pytest tests/test_logic.py

Running Specific Test Functions

pytest -k test_function_name

Checking Test Coverage

pytest --cov=bq_mcp_server

Local Development

Starting the MCP Server

uv run bq_mcp_server

Starting the FastAPI REST API Server

uvicorn bq_mcp_server.adapters.web:app --reload

Development Commands

Code Formatting and Linting

# Code formatting
ruff format

# Linting checks
ruff check

# Automatic fixes
ruff check --fix

Dependency Management

# Adding new dependencies
uv add <package>

# Adding development dependencies
uv add --dev <package>

# Updating dependencies
uv sync