Creativity · MCP — server

MCP BigQuery Server

The MCP BigQuery server lets Claude and other MCP clients list datasets, describe tables, and run SQL against a BigQuery project. Google hasn't shipped an official Anthropic-style MCP server yet (as of April 2026), but several solid community implementations exist that wrap the `google-cloud-bigquery` client library. Pair it with a dedicated service account so the LLM can't exceed its intended scope.

MCP facts

Kind: server
Ecosystem: anthropic-mcp
Language: Python
Transports: stdio

Capabilities

Tools: list_datasets, list_tables, describe_table, run_query
Resources: dataset and table schemas as MCP resources
Auth: GOOGLE_APPLICATION_CREDENTIALS service-account JSON

Install

uvx mcp-server-bigquery

Configuration

{
  "mcpServers": {
    "bigquery": {
      "command": "uvx",
      "args": ["mcp-server-bigquery", "--project", "my-gcp-project"],
      "env": {
        "GOOGLE_APPLICATION_CREDENTIALS": "/Users/you/.gcp/mcp-sa.json"
      }
    }
  }
}

Frequently asked questions

Does Google ship an official BigQuery MCP server?

Not as of April 2026. The common route is a community package (e.g. mcp-server-bigquery) built on top of google-cloud-bigquery. Check release cadence before adopting.

How do I keep BigQuery costs bounded?

Give the service account the roles/bigquery.dataViewer role on only the datasets you want exposed, set a query `maximumBytesBilled` cap, and consider a BI Engine reservation for frequent exploration.

Can I run ML.PREDICT via the server?

Yes — it's just SQL. Whether the service account can call Vertex AI–backed functions depends on IAM; usually you'll add roles/aiplatform.user.

Sources

Google Cloud BigQuery docs — accessed 2026-04-20
Model Context Protocol specification — accessed 2026-04-20