Skip to main content

Quick Overview

This interview features Google Cloud Developer Relations Engineers Chrissie Goodrich and Mike DeAngelo discussing the Looker command-line interface. DeAngelo details the project's background, its transition from the legacy Gazer tool, and its recent architectural updates. The discussion focuses on how Looker CLI enables both DevOps engineering automation and autonomous AI agent integration.

Key Points

  • 1.Looker CLI is a command-line program originally created as Gazer in Ruby to migrate looks and dashboards between instances, now completely rewritten in Go.
  • 2.The tool exposes the full Looker API surface while providing bespoke administrative and DevOps utilities such as instance provisioning and project management.
  • 3.Unlike Model Context Protocol servers that flood LLM context windows with tool definitions, the CLI lets agents discover and execute commands on demand without upfront context bloat.
  • 4.Built-in discovery commands like meta tree and meta search allow both human operators and AI agents to explore commands and generate JSON body templates for API requests.
  • 5.The health pulse command integrates functionality from the open source Henry tool to audit database connections, explore performance, query latency, and scheduled task failures.

Summary

Developer Relations Engineers Chrissie Goodrich and Mike DeAngelo discuss the history, design, and practical applications of Looker CLI. DeAngelo recounts that the utility originated roughly eight years ago under the name Gazer. Written in Ruby, Gazer solved a foundational operational headache: exporting dashboards and looks as JSON documents and recreating them across different environments, such as promoting assets from development to production. Over time, Looker DevOps required broader automation capabilities for creating users, provisioning projects, and managing connections. Because the broader software ecosystem shifted away from Ruby, DeAngelo, Josh Temple, and Victor Hoys executed a complete rewrite in Go, expanding the program to expose Looker's entire API surface alongside specialized DevOps commands.

The conversation examines how Looker CLI complements AI agent development in contrast to Model Context Protocol, commonly known as MCP. Goodrich and DeAngelo note that MCP servers populate an agent's context window with every tool description upfront. This design creates practical constraints, as context saturation typically limits an MCP implementation to roughly fifty tools. Looker CLI avoids context pollution because no commands or documentation reside in the context window until queried. Supported by Looker skills and built-in discovery utilities, an agent can dynamically explore the command hierarchy and invoke specific operations on demand.

DeAngelo demonstrates terminal interaction with the CLI, illustrating its subcommands and discovery mechanisms. Running looker-cli meta reveals available commands, while looker-cli meta tree displays an exhaustive structural map of all endpoints, including permissions, users, queries, and connections. When looking for specific functionality, looker-cli meta search filtered by a keyword such as dashboard outputs all relevant subcommands. DeAngelo highlights a template switch on API calls that generates empty JSON bodies. Language models excel at filling out predefined JSON templates, eliminating the need to write custom glue code to orchestrate multi-step API sequences.

To showcase real-world agent orchestration, DeAngelo launches the Antigravity CLI agent using Gemini Pro. He instructs the agent in natural language to scan his personal Looker home folder for dashboards and looks starting with the prefix Test and delete them. The agent interprets the request, executes looker-cli discovery commands, identifies Dashboard 4842 and Look 6103, and removes them from the instance. Next, the agent executes the looker-cli health pulse command, a feature brought over from the open source diagnostic project Henry. The tool runs an automated audit across database connections, instance average query runtimes, dashboard errors, explore performance, and scheduled delivery failures. DeAngelo concludes by noting that Looker CLI receives regular updates aligning with monthly Looker platform releases.

Evolution from Gazer to Looker CLI

Mike DeAngelo explains that Looker CLI began approximately eight years ago as Gazer, a Ruby tool designed to migrate dashboards and looks between development and production environments. Written in collaboration with Josh Temple and Victor Hoys, the tool was rewritten in Go to eliminate Ruby dependencies and support modern agentic workflows. Beyond content migration, the modernized CLI covers the entire Looker API surface alongside dedicated utilities for managing users, projects, and data sources.

Comparing CLI Workflows with MCP Servers

Model Context Protocol servers typically struggle with extensive API surfaces because tool descriptions consume substantial context window space, limiting servers to around fifty tools. In contrast, Looker CLI does not inject tool definitions into the context window upfront. AI agents can dynamically query help screens, inspect command trees, and discover subcommands as needed, enabling complete API access without overloading the language model.

Command Discovery and JSON Templates

Looker CLI provides introspection utilities including meta tree and meta search to navigate its extensive hierarchy. By querying keywords such as dashboard, users and agents can locate all matching commands and endpoints. Additionally, an API template switch outputs blank JSON schema structures, which language models can easily parse and populate rather than having to construct complex payload bodies from scratch.

Agent Automation and Instance Health Auditing

In a live terminal demonstration, the Antigravity agent uses Looker CLI to search a user home folder for test dashboards and looks, returning specific asset IDs before deleting them. The demonstration also highlights the health pulse command, adapted from the open source Henry diagnostic project. This tool audits active database connections, query response times, dashboard errors, explore performance, and failed scheduled jobs in a single execution.

The Bottom Line

The video establishes Looker CLI as both a comprehensive DevOps automation utility and an efficient, context-friendly interface for AI agents. By providing full API coverage, discovery commands, JSON templating, and integrated health diagnostics, it replaces manual scripting across Looker environments. While it demonstrates strong integration with tools like Antigravity and Gemini, the discussion notes that the CLI functions identically across other agent frameworks without prescribing a single orchestration method.

FAQ

What is Looker CLI and what does the command-line utility do?

Looker CLI is an open source command-line tool written in Go that provides full access to the Looker API surface along with specialized DevOps utilities. It enables administrators and developers to manage users, deploy projects, migrate dashboards, and monitor instance health either manually or through automated scripts.

What original problem was the Gazer Ruby tool built to solve?

Gazer was created roughly eight years ago in Ruby to export dashboards and looks as JSON files and recreate them on different Looker instances, such as moving assets between development and production environments.

Why is Looker CLI better suited for broad API coverage than MCP servers?

Model Context Protocol servers inject tool descriptions into the prompt context, creating context window limits that typically cap usable tools around fifty. Looker CLI does not load tools into context upfront, allowing agents to explore hundreds of commands dynamically as needed.

How does the template flag in Looker CLI help AI agents construct API requests?

The template flag outputs an empty JSON document representing the expected API request body. Language models can reliably ingest this template and fill out required parameters instead of having to build complex JSON structures from scratch.

What diagnostics does the Looker CLI health pulse command run on an instance?

Incorporated from the open source tool Henry, health pulse audits database connections, instance query latency, dashboard errors, explore query performance, and scheduled plan delivery failures.

Worth watching for

Looker administrators, DevOps engineers, and software developers looking to automate Looker management or integrate Looker API operations into AI agent workflows.

  • looker
  • looker-cli
  • google-cloud
  • ai-agents
  • devops
  • mcp