Summary
A warehouse an agent can be pointed at without a conversation about permissions first.
The entire surface is one read-only query tool, which is what makes it easy to hand to a model: there is no write, no load and no schema-change path to fence off. What it does not carry is discovery — nothing lists datasets, tables or columns — so the schema has to come from you or from a query the agent already knows how to write.
What it is
A single-tool server over BigQuery that runs read-only SQL and returns the result.
What you get
- Read-only BigQuery SQL run from the conversation, with the rows returned into it
- The project fixed when the server starts, so each call carries only its sql
- A single entry point, which makes what the agent can do here exactly what a read-only SQL user can do
Requirements
A BigQuery project and credentials on the machine that can read it. The project id is supplied when the server starts rather than per query.
Setup effort
One command plus a key — npx -y @ergut/mcp-bigquery-server --project-id your-project-id, then supply credentials