Navigating AI automation in Looker: Essential developer tools for agents

Summary

1. 1.Looker Skills. Looker Skills are Markdown files that serve as structured instruction manuals for AI agents, allowing both human developers and agents to follow consistent LookML development steps. To maintain a clean context window, an agent reads only a minimal skill manifest that exposes the title and description from the header of each file. When a user prompt requires a specific task, the agent evaluates intent against the manifest and retrieves the complete instructions into working memory. This replaces general model assumptions with deterministic, domain-specific rules. For large projects, a primary agent can spawn temporary subagents containing only the specific skills needed for their subtask, keeping conversation threads organized. Predefined skills in Looker open-source repositories support onboarding, auditing, and performance optimization.
2. 2.Looker Command Line Interface. The Looker Command Line Interface, or Looker CLI, is a Go-based tool that wraps Looker API 4.0 to let agents automate administrative operations through terminal commands. Installed on a developer machine, the tool executes tasks directly against Looker, such as updating schedules, managing folders, administering users, and modifying dashboards. Authentication is handled using standard API credentials or OAuth. Because Looker API credentials inherit 100 percent of the permissions and folder access belonging to the associated user account, using personal credentials presents an excess privilege risk. The video recommends establishing dedicated service accounts with least privilege to create an isolated security boundary and a reliable audit trail for all automated CLI actions.
3. 3.Looker Managed MCP Server. The Model Context Protocol replaces raw shell scripts with a curated, schema-validated toolset that agents can execute safely. Looker provides a native MCP server hosted directly within the platform, requiring zero external hosting infrastructure. Within the platform administration panel, administrators can toggle access to specific tools, such as running queries or building dashboard elements. External AI clients, including Antigravity CLI, Cursor, and Claude Desktop, connect to this server using standard API credentials or OAuth. Because tool activation on the managed server applies instance-wide to all connecting clients, developers must still enforce individual user boundaries through assigned Looker user roles, model sets, and folder permissions.
4. 4.MCP Toolbox for Databases. The open-source MCP Toolbox for Databases provides a self-hosted alternative to Looker's built-in managed server. Deployed locally, in Docker, or across cloud infrastructure, it relies on a YAML configuration file to define data connections. This configuration enables multi-source orchestration, allowing an agent to query Looker alongside native databases like BigQuery and AlloyDB within a single session. Because developers configure the underlying YAML files directly, this toolbox provides client-level scoping to expose distinct, restricted toolsets to specific AI agents.
5. 5.Looker Visual Studio Code Extension. The Looker Visual Studio Code extension brings LookML development workflows and agent integrations directly into the developer editor. It synchronizes local project edits with Looker development mode while maintaining local Git version control. The extension includes a step-by-step onboarding wizard to configure API or OAuth credentials, pull workspace files, install Looker Skills, and run an internal proxy that lets local AI agents communicate directly with Looker's managed MCP server under the active authenticated session.