BigQuery MCP Server
What it is: A community-maintained MCP server that exposes BigQuery datasets, table schemas, and ad-hoc SQL execution to any MCP-compatible agent. Auth flows through your normal Application Default Credentials.
Why it matters for data work
BigQuery is where most modern analytics lives. Giving an agent direct, read-friendly access to your warehouse — with the agent able to inspect INFORMATION_SCHEMA on demand — turns ad-hoc analysis into a chat conversation instead of a notebook session.
Install & configure
uvx mcp-server-bigquery --project YOUR_GCP_PROJECT --location USAdd to your ~/.claude.json MCP servers block, set GOOGLE_APPLICATION_CREDENTIALS, and restart Claude Code. Tools include list_datasets, list_tables, describe_table, and execute_query.
Example usage
Ask "break down conversions by acquisition channel for Q1 and Q2 — use the events_* tables, sample 10% to keep cost under a dollar". The agent will introspect the table schema, write a sampled query with the right partition filter, and present results inline.
Author & links
Author: Lucas Hild (community maintainer)
Repo: github.com/LucasHild/mcp-server-bigquery
License: MIT
Cost watch: An agent will happily fire off scans against TB-scale tables. Always set a billing budget alert and consider running with a service account that has limited bytes-billed quota when experimenting.