mcp-server-bigquery

What it does

This MCP server bridges Claude to BigQuery, enabling language models to execute queries, list tables, and inspect table schemas. It implements three tools: execute-query (runs BigQuery SQL), list-tables (enumerates tables across datasets), and describe-table (shows schema and column details). The server connects to BigQuery using either a service account key file or ambient GCP credentials on the host.

Who it's for

Data engineers and analysts who want Claude to interact directly with BigQuery from Claude Code, a Python script, or Claude Desktop. Useful when you need an LLM to explore schemas, draft queries, or answer ad-hoc analytical questions over live data.

Common use cases

• Query a BigQuery dataset directly from Claude Code to answer analytical questions
• Have Claude explore table schemas and suggest query optimizations
• Let an LLM generate and execute ad-hoc SQL against production data
• Scope Claude's access to specific datasets via --dataset filters

Setup pitfalls

• Both BIGQUERY_PROJECT and BIGQUERY_LOCATION environment variables are required; without them the server will not start.
• GCP credentials must be available: either via a service account key file (set BIGQUERY_KEY_FILE) or ambient credentials (e.g., gcloud auth application-default login on the host). Misconfigured credentials will fail silently at query time.
• By default, all datasets in the project are exposed. To limit Claude's access, use --dataset my_dataset_1 --dataset my_dataset_2 or the comma-separated BIGQUERY_DATASETS variable.
• Long-running queries hang by default; set BIGQUERY_TIMEOUT (in seconds) to avoid blocking indefinitely.