Agentic analytics with the Data Agent Kit

# What the Data Agent Kit does

The kit is available as an extension for certain VS Code forks and as a plugin for other tools, so you don't need to leave your development environment to start an investigation.

# Core mechanisms

• Model Context Protocol (MCP): an open standard that connects the agent to tools, databases, and cloud services.
• Skills: markdown files that teach an agent how to use your specific stack and tools.

# How a real investigation looks (average order value example)

Scenario: A manager asks why average order value (AOV) dropped 7% in January while revenue stayed flat.

1. 1Start in the IDE chat. Ask the agent to calculate monthly AOV for a date range using your warehouse tables. The agent prepares queries and asks for permission to run them.

1. 1Permission control. Before executing, the IDE prompts you to approve the MCP tool calls. You can allow once for an audit trail or choose "always allow" to streamline repeated workflows.

1. 1Drill down. The agent reports AOV was stable through December and fell in January. You ask it to break down January by order type. The agent finds a new channel labeled B2B-Wholesale with a much lower AOV (~$75) while Online and Offline remained ~ $110.

1. 1Cross-system checks. To learn who these wholesale buyers are, the agent switches to the Cloud SQL MCP and queries the operational customers table. It finds all B2B accounts are new business entities created in the last 30 days.

1. 1Metadata lookup. The agent inspects campaign files in Cloud Storage and finds a promo code (BIGORDER25) that applied a 25% discount to many B2B orders. This explains the blended AOV drop while total revenue stayed flat.

1. 1Summary. In a single chat session the agent queried analytical data (BigQuery), operational records (Cloud SQL), and unstructured metadata (Cloud Storage), then produced an executive summary you can send to the director.

# Practical governance and workflow notes

• Permission gating: every execution requires permission, so you retain control and an audit trail.
• Inspectable outputs: agents provide an execution trail showing tool calls and raw SQL for review.
• Reuse: skills are markdown files you can create and version to teach the agent about your schemas, conventions, and workflows.

# When to use it

Use the Data Agent Kit when you need to answer open-ended, cross-system questions that would otherwise require writing the same queries repeatedly across multiple consoles. It reduces context switching and accelerates root-cause analysis while preserving manual review points.