Google Cloud Releases Plugin Connecting Coding Agents to the Cloud

On September 11, 2026, Google Cloud released a new plugin for AI coding agents designed to streamline how tools like Claude Code, Codex CLI, and Antigravity CLI work with Google Cloud resources. The google-cloud-developer plugin bundles agent skills and Model Context Protocol servers into a single package, removing the need to manually configure separate wrappers for each assistant.

The Tool Coupling Problem

In a Google Cloud blog post written by Jonathan Lee, the company describes a phenomenon it calls the "tool coupling problem." Individual agent skills are useful because they're lightweight on context window usage and easy to install, but managing them one by one becomes cumbersome, especially when they work best paired with other skills or MCP servers serving a shared goal. Plugins are meant to solve this by packaging domain knowledge, workflow recommendations, and tools for interacting with real environments into coherent, installable units.

What the Flagship Plugin Does

The google-cloud-developer plugin, published in the Google Agent Skills repository, helps agents navigate the basics of Google Cloud: authentication and authorization, project management, and applying safeguards to operations run through the gcloud CLI. It also bundles configuration for the Developer Knowledge MCP server, meant to give agents access to current, official Google developer documentation.

Google's post walks through an example use case: a user asks the agent to set up an account and its first billing-enabled project, then to authenticate the local machine so a script can call APIs as a service identity rather than as the user themselves. According to Google, an agent equipped with the plugin first quietly checks the environment (CLI availability, existing projects and organizations), then applies IAM best practices to avoid risks such as accidental key leaks or keys ending up in a commit, and finally presents a plan of action and asks for approval before modifying any resources.

An Open Standard Instead of Separate Wrappers

The plugin is built to the Agent Plugins specification, which Google describes as an open, vendor-neutral standard for packaging agent skills and MCP servers into portable, mutually compatible units. Instead of maintaining separate configurations and wrappers for each AI assistant, the standard provides a unified manifest and directory structure. Every plugin in the Google Agent Skills repository is meant to follow the same standard, ensuring consistent tool access regardless of which coding environment a developer uses.

Installation Across Three Environments

Google has published installation instructions for all three tools. In Antigravity CLI, the plugin is installed with the agy plugin install command pointing to the google/skills/plugins/cloud/google-cloud-developer repository. In Claude Code, users first add the marketplace with claude plugin marketplace add google/skills, then install the plugin with claude plugin install google-cloud-developer@google-plugins. A similar two-step process, based on the codex plugin marketplace add and codex plugin add commands, applies to Codex CLI. Google has also published a dedicated codelab walking through plugin setup in Antigravity.

The release fits a broader trend of major tech companies embedding AI agents directly into developer workflows rather than confining them to chat interactions. The google-cloud-developer plugin complements Google's existing AI-assisted coding tools, such as Gemini Code Assist and Gemini CLI, expanding the company's product ecosystem for developers working with Google Cloud.

For teams using multiple coding assistants at once, the main benefit is configuration consistency. Instead of separately tuning access to Google Cloud resources for each tool, a developer installs a single plugin conforming to the same manifest, regardless of whether they're working in Antigravity, Claude Code, or Codex CLI. That lowers the barrier for teams just getting started with Google Cloud and reduces the risk of configuration drift between different environments.