BigQuery MCP server
Google's MCP Toolbox in its prebuilt BigQuery mode: explore datasets, run SQL, forecast, and ask data-insight questions over a warehouse.
BigQuery support ships inside Google's MCP Toolbox for Databases, an open-source MCP server maintained under the googleapis organization. Run the toolbox binary with --prebuilt bigquery and it instantly exposes a curated set of BigQuery tools without writing any tool definitions, so an agent can list datasets and tables, inspect their schemas, run SQL, forecast time series, perform key-driver contribution analysis, and search a data catalog in natural language. The prebuilt set is the fast path; the same toolbox can also load hand-authored YAML tool configs when you want tighter control over exactly which queries an agent may run.
The server connects with Application Default Credentials, so it inherits whatever IAM identity you have configured through gcloud, and it scopes all work to the project named in the BIGQUERY_PROJECT environment variable. It is distributed as a single Go binary (with Homebrew, container image, and go install options), runs over stdio for desktop clients via the --stdio flag, and can also serve HTTP for networked deployments. Because it leans on BigQuery's own access controls, the agent only ever sees data the underlying credential is permitted to read.
Quick install
Copy-paste configs are provided for all 8 supported clients. Pick your client below.
Add to ~/.claude.json
{
"mcpServers": {
"bigquery": {
"command": "docker",
"args": [
"run",
"-i",
"--rm",
"-e",
"BIGQUERY_PROJECT",
"us-central1-docker.pkg.dev/database-toolbox/toolbox/toolbox:latest",
"--prebuilt",
"bigquery",
"--stdio"
],
"env": {
"BIGQUERY_PROJECT": "<BIGQUERY_PROJECT>"
}
}
}
}claude mcp add bigquery -- docker run -i --rm -e BIGQUERY_PROJECT us-central1-docker.pkg.dev/database-toolbox/toolbox/toolbox:latest --prebuilt bigquery --stdioAvailable tools
| Tool | Description |
|---|---|
| execute_sql | Executes an arbitrary SQL statement against BigQuery and returns the result rows. |
| list_dataset_ids | Lists the dataset IDs available in the configured project. |
| list_table_ids | Lists the table IDs within a given BigQuery dataset. |
| get_dataset_info | Retrieves metadata for a BigQuery dataset. |
| get_table_info | Retrieves schema and metadata for a BigQuery table. |
| forecast | Forecasts time-series data using BigQuery's built-in forecasting (AI.FORECAST). |
| analyze_contribution | Performs contribution analysis (key-driver analysis) to explain changes in a metric. |
| search_catalog | Searches for tables and other entries in the data catalog using a natural-language query. |
| ask_data_insights | Answers natural-language questions about the contents of tables by performing data analysis. |
Required configuration
- BIGQUERY_PROJECTRequired
Default Google Cloud project ID used for all BigQuery operations.
- GOOGLE_APPLICATION_CREDENTIALSOptional
Path to a service-account key file for Application Default Credentials. Optional if ADC is already configured via gcloud.
What you can do with it
Conversational warehouse exploration
Let an analyst-style agent list datasets and tables, read their schemas, and run SQL to answer ad-hoc questions over BigQuery without leaving the chat.
Forecasting and driver analysis
Use the forecast and analyze_contribution tools so the agent can project a metric forward and explain which dimensions drove a change, all backed by BigQuery's native ML functions.
FAQ
- Is it free?
- The MCP Toolbox is open source and free to run. You pay only BigQuery's normal storage and query (bytes-scanned) costs for the work the agent triggers.
- Does it support remote/OAuth?
- It runs locally over stdio (or HTTP for networked use) and authenticates with Google Application Default Credentials rather than a per-tool OAuth flow. The credential's IAM grants determine what the agent can access.