Google ADK and Looker: MCP agent setup tutorial

Chrissie Goodrich walks through a basic Google Agent Development Kit (ADK) setup that connects an AI agent to Looker through MCP Toolbox for Databases. The tutorial runs a local application grounded in enterprise data. It does not establish that the language model itself runs locally or that the setup works offline.

The walkthrough starts with a Looker client ID and client secret, stored as environment variables. Users who cannot manage API keys may need help from their administrator. Goodrich then creates a Python virtual environment with uv, installs Google ADK and additional tools, and generates the agent files with ADK create.

The tools.yaml configuration defines a Looker source with its base URL, an instance port if required, and references to the credential environment variables. The first tool, get_connections, fetches database metadata. Its description helps the model decide when to call it.

The Python agent configuration connects to a local MCP server through standard input and output and assigns a tool set to a root agent. Goodrich tests tool discovery with ADK web, then adds another Looker tool and restarts the app to show the expanded selection. The demonstration focuses on configuration and available tools rather than model performance.