Google provides two official ways to connect AI assistants to BigQuery via MCP: a managed Remote Server and an open-source self-hosted Toolbox. This hub covers the decision framework, setup procedures, authentication, custom query patterns, and cost management across six notes. Prerequisites: familiarity with MCP fundamentals and basic GCP/BigQuery usage.
Prerequisites
- A Google Cloud project with BigQuery enabled
gcloudCLI installed and authenticated- An MCP client: Claude Desktop, Claude Code, or similar
- Familiarity with MCP concepts (clients, servers, tools)
Reading Order
Choosing Between BigQuery MCP Options — decision framework for selecting between options. Claude Desktop users can start with the Remote Server; Claude Code users should use the Toolbox; for quick ad-hoc queries, the bq CLI may suffice.
BigQuery Remote MCP Server Setup — enabling the managed service, configuring the client, and the token expiration limitation that makes it impractical for Claude Code.
GCP Application Default Credentials — the authentication mechanism the Toolbox uses. The most common stumbling block in setup; read before or alongside the Toolbox setup note.
BigQuery MCP Toolbox Setup — installing the binary, configuring authentication, and setting up both Claude Desktop and Claude Code.
Custom Parameterized MCP Queries — defining which queries the AI can run, with which parameters, via a tools.yaml that can be shared and committed.
AI Query Cost Control for BigQuery MCP — cost and safety implications of giving an AI direct query access to BigQuery, with mitigation strategies.
Related Fundamentals
- MCP Protocol Architecture — How MCP works at the protocol level
- BigQuery IAM Patterns — The role-based access control that underpins both MCP options
- BigQuery Cost Model — How BigQuery pricing works, independent of MCP
- Security Posture for AI Agents — Broader security principles for AI tools accessing data