@renzo.johnson/bigquery-mcp
v26.11.15
Published
BigQuery MCP server for Claude Code, Cursor, Copilot and five more AI agents. mcprush.com-45569233d2e038a123849647
Maintainers
Readme
BigQuery MCP
BigQuery MCP server for Claude Code, Cursor, Copilot and five more AI agents. This is the MCP server from the BigQuery MCP VS Code extension, packaged to run with npx in any MCP client.
Requires Node.js 22 or newer.
npx -y @renzo.johnson/bigquery-mcpThe server speaks MCP over stdio, so you normally let your AI agent start it, as below.
Claude Code
claude mcp add bigquery-mcp -- npx -y @renzo.johnson/bigquery-mcpOr add it to .mcp.json in your project (or ~/.claude.json for every project):
{
"mcpServers": {
"bigquery-mcp": {
"command": "npx",
"args": ["-y", "@renzo.johnson/bigquery-mcp"],
"env": {
"GOOGLE_APPLICATION_CREDENTIALS": "/path/to/service-account.json"
}
}
}
}Cursor
Add the same block to ~/.cursor/mcp.json (or .cursor/mcp.json in a project):
{
"mcpServers": {
"bigquery-mcp": {
"command": "npx",
"args": ["-y", "@renzo.johnson/bigquery-mcp"],
"env": {
"GOOGLE_APPLICATION_CREDENTIALS": "/path/to/service-account.json"
}
}
}
}Credentials
The server signs in to Google in this order:
| Option | Environment | Best for |
|--------|-------------|----------|
| OAuth user | GOOGLE_OAUTH_CLIENT_ID, GOOGLE_OAUTH_CLIENT_SECRET, GOOGLE_OAUTH_REFRESH_TOKEN (all three) | Personal Google accounts |
| Service account or saved credentials | GOOGLE_APPLICATION_CREDENTIALS = path to a JSON key | Servers, Workspace admins |
| gcloud | none: run gcloud auth application-default login once | Developers with the Google Cloud CLI |
Setting only some of the three OAuth variables stops the server with an error that names the missing ones. Enable the BigQuery API in your Google Cloud project.
Optional settings:
| Variable | What it does |
|----------|--------------|
| BIGQUERY_MCP_PROJECT_ID | Google Cloud project ID used when a request does not name one |
| BIGQUERY_MCP_MAX_BYTES_BILLED | Largest number of bytes a query may scan (default 1000000000) |
Set bigqueryMcp.projectId to the project that runs your queries. bq_query runs only SELECT statements and dry-runs each one first; a query that would scan more than bigqueryMcp.maxBytesBilled (1 GB by default) is refused instead of billed.
Available Tools (5)
| Tool | What It Does | Example Prompt |
|------|-------------|----------------|
| bq_list_datasets | Datasets in a project | "List my datasets" |
| bq_list_tables | Tables and views in a dataset | "What tables are in analytics?" |
| bq_get_table | Schema, row count and size | "Show the schema of analytics.events" |
| bq_query | Run a SELECT, after a dry-run cost check | "Orders per day last week" |
Safety
BigQuery MCP runs SELECT queries only, with a byte limit. Every query is dry-run first. Anything other than a single SELECT, or a query that would scan more than BIGQUERY_MCP_MAX_BYTES_BILLED (1 GB by default), is refused before it runs or bills.
VS Code
Prefer a setup wizard that registers the server with eight agents for you? Install the BigQuery MCP extension instead (also on the VS Code Marketplace). Use one or the other, not both.
