Data Agent Kit

Data Agent Kit is a free plugin that connects your coding agent to 15+ Google Data Cloud services, including BigQuery, Cloud Storage, Spanner, AlloyDB, Managed Service for Apache Spark, and Knowledge Catalog.

Spend your time on insights, not boilerplate. Describe a data task in plain language, and watch your agent handle the heavy lifting for you: writing SQL queries, building notebooks, generating pipeline code, setting up ML workflows, and more.

• A supported coding agent: Antigravity CLI, Claude Code, Codex CLI, or another Agent Plugins–compatible client.

• Node.js (latest LTS recommended).

• Google Cloud CLI (gcloud), installed and authenticated:

  1. Install the gcloud CLI.

  2. Log in with your Google Cloud account:

     gcloud auth login

  3. Log in to Application Default Credentials (ADC):

     gcloud auth application-default login

agy plugin install https://github.com/GoogleCloudPlatform/data-agent-kit-plugin

claude plugin install data-agent-kit-starter-pack@claude-plugins-official

codex plugin marketplace add GoogleCloudPlatform/data-agent-kit-plugin
codex plugin add dak@dak-marketplace

https://github.com/GoogleCloudPlatform/data-agent-kit-plugin

/dak:dak-setup

codex -s danger-full-access

$dak:dak-setup

• Data engineering: Build pipelines that turn source data into analysis-ready tables, with quality checks along the way. Develop transformations with dbt or Dataform, schedule workflows with Managed Service for Apache Airflow, and debug failed runs by tracing execution logs and code.
• Data science: Develop experiments in Jupyter notebooks on local Python or remote Managed Service for Apache Spark kernels. Prepare features from BigQuery, train models with BigQuery ML, and schedule batch inference.
• Data analysis: Explore data across BigQuery, AlloyDB, and Cloud SQL. Find datasets with Knowledge Catalog, guided by data quality metrics and lineage, then build visualizations or reusable data models.

• "Create a PySpark notebook to train a distributed Random Forest model on the BigQuery table my_transactions
• "Find datasets about customer orders in Knowledge Catalog and show their lineage"
• "Create an Airflow DAG that runs the dedup notebook, the dbt pipeline, and model training every Monday morning.
• "My last Dataflow job failed. Find the root cause in the logs and propose a fix"

If you see could not find default credentials or other auth errors, ensure you are logged into gcloud.

Exit your agent, run gcloud auth login followed by gcloud auth application-default login, then restart your agent.

Your agent can run tools and commands on your behalf. Apply the principle of least privilege to every CLI, MCP server, and resource it can access:

• Use service account impersonation instead of end-user credentials.
• Grant the service account only the roles it needs.
• Use Principal Access Boundary policies with a condition in the policy binding to restrict your agent's service accounts to intended projects.

Learn how to mitigate prompt injection risks with Google Cloud MCP.

Data Agent Kit collects usage statistics (such as when included skills and MCP tools are used) to improve reliability and performance. No user code, file contents, or application data values are collected.

To opt out, either set DO_NOT_TRACK=1 in your environment:

export DO_NOT_TRACK=1

Or create or update ~/.data_agent_kit/config.json:

{
  "enableTelemetry": false
}

We want your feedback! Please open an issue to report bugs, file feature requests, or ask questions.

• Data Agent Kit homepage
• Data Agent Kit documentation