select-bq

A SELECT-only BigQuery CLI wrapper for agentic use (e.g. with Cursor). Wraps the official bq CLI and enforces:

• SELECT-only: Queries are validated via SQL AST parsing—no DML, DDL, scripting, EXECUTE IMMEDIATE, or hidden CTEs that could modify data.
• Allowlist: Optional config restricts queries to allowed tables (string patterns project.dataset.table). Empty or missing allowlist = no restriction.
• Audit logging: All queries (including rejections) are logged to a YAML file with timestamps.

Install

pip install select-bq

Requires the Google Cloud SDK (bq CLI) and gcloud auth login.

Setup

Create a config file .select-bq.yaml in your project root:

# Where to log queries (default: select-bq-queries.yaml in current dir)
log_path: select-bq-queries.yaml

# Optional allowlist. Omit or leave empty to allow all tables.
# When present, only these patterns can be queried (three dot-separated parts).
allowlist:
  - my-gcp-project.analytics.events
  - my-gcp-project.analytics.*           # all tables in this dataset
  - my-gcp-project.*.*                 # entire project
  - my-gcp-project.analytics.onecrm-*  # glob on table name (fnmatch)

• log_path — Path for the query log (default: select-bq-queries.yaml). Use an absolute path to log outside the project.
• allowlist — List of strings project.dataset.table. Use * for a full segment wildcard; use prefix* / *suffix style globs in a segment when you need pattern matching. Omit to allow all tables.

To use a different config path: select-bq query --config ./my-config.yaml "SELECT 1".

To use an external allowlist file:

log_path: select-bq-queries.yaml
allowlist_path: allowlist.yaml

Usage

# Run a SELECT query (same as bq query, but validated)
select-bq query "SELECT 1"
select-bq query "SELECT * FROM project.dataset.table LIMIT 10" --format=pretty

# Query from file
select-bq query -f query.sql

# Custom config
select-bq query --config ./my-config.yaml "SELECT * FROM my_table"

# Use Standard SQL (default) or legacy SQL
select-bq query --use_legacy_sql=false "SELECT 1"
select-bq query --use_legacy_sql=true "SELECT 1"

# All bq query flags are passed through (format, project_id, etc.)
select-bq query --format=pretty --project_id=my-project "SELECT 1"

Query Log

Logged to log_path (default select-bq-queries.yaml):

queries:
  - timestamp: "2025-03-09T12:00:00.000000+00:00"
    query: "SELECT 1"
    success: true
  - timestamp: "2025-03-09T12:01:00.000000+00:00"
    query: "INSERT INTO t VALUES (1)"
    success: false
    error: "Statement type 'Insert' is not allowed. Only SELECT queries are permitted."

Security

• AST parsing: Uses sqlglot with BigQuery dialect. Only SELECT statements are allowed; INSERT, UPDATE, DELETE, CREATE, DROP, EXECUTE IMMEDIATE, DECLARE, SET, and similar are rejected.
• Allowlist: When configured, only tables matching the allowlist patterns can be queried (e.g. dataset.* for all tables in a dataset).
• No eval/exec: Validation is purely structural—no dynamic execution of user input.

Cursor Integration

Add to your project's Cursor rules or AGENTS.md:

Use `select-bq query "SELECT ..."` when querying BigQuery. Do not use raw `bq` for queries.

Publishing

To build and publish to PyPI:

pip install build twine
python -m build
twine upload dist/*