Summary
Tables arrive as resources, not just as something a tool call has to discover.
Pointing --dataset at a dataset registers a resource per table plus its schema, so a client can browse what exists before it writes any SQL. If you would rather the model not wander, both discovery tools have off switches and query stays registered on its own.
What it is
An MCP server for Google BigQuery. It exposes tables and their schemas as resources and runs SQL against your project, so an agent can explore a dataset and then query it.
What you get
- query(sql) — execute BigQuery SQL and return the results
- list_tables and get_schema, both taking an optional table argument
- Resources: bigquery://tables, and bigquery://tables/{table}/schema for one table
- --dataset registers resources for every table in a dataset; --table scopes to named tables
- Both discovery tools can be switched off, leaving only query registered
- Stdio, sse or streamable-http transports, with --port defaulting to 8000
Requirements
A Google Cloud project with BigQuery enabled and permission to submit jobs, plus the gcloud CLI and uv. Authenticate once with gcloud auth application-default login, then run uvx sl-bigquery-mcp with --dataset and --project. Version 0.1.9.
Setup effort
One command plus a key — uvx sl-bigquery-mcp, then supply credentials