- Centralize Google Cloud capabilities into installable packages that support open standards.
- Integrate official documentation in real time using MCP servers to avoid AI hallucinations.
- Implement IAM security and validation layers before making changes to the infrastructure.
- It supports native integration with tools such as Claude Code, Codex, and Antigravity CLI.
If you work in the development world and use AI agents to write code, you know that getting the machine to understand exactly how the cloud infrastructure works can sometimes be a real headache. Until now, we had to provide manual documentation or configure remote MCP servers, but things became unmanageable when we needed multiple tools to work together to solve a complex problem. To solve this problem, Google Cloud has launched the AI Agents plugin , a solution designed to package skills and tools into a single installable package.
The great advantage of this system is that it prevents developers from having to manage complex dependencies or different configurations for each assistant. By being based on an open and neutral standard , it allows the agent to access in-depth knowledge of specific products without cluttering the context window, making interaction with the Google ecosystem much smoother and, above all, more effective in real-world work environments.
What exactly is the google-cloud-developer plugin?
This plugin is essentially a survival kit for any AI agent that needs to operate on Google Cloud. Its primary purpose is to help the assistant navigate the platform's fundamentals , covering critical topics such as project management, authorization, authentication, and, very importantly, implementing security barriers when executing gcloud CLI commands.
One of its crown jewels is its integration of the Developer Knowledge MCP server configuration. Thanks to this, the agent doesn't make things up; instead, it has a real-time anchor to Google's official documentation, ensuring that responses are up-to-date and not based on outdated data.
Types of plugins and how they work
Google has structured its skills repository to allow for orderly growth, dividing add-ons into two main blocks:
- Foundational plugins: These are the basics for everyone. They focus on project setup, architectural design, and the ability to find relevant documentation.
- Plugins focused on domains: These are more specialized and offer best technical practices for specific areas within the Google cloud.
To prevent the agent from getting lost, there is a route map called google-cloud-discovery.mdThis mechanism acts as a guide, redirecting the agent towards deeper capabilities (such as those of GKE or Google Cloud's WAF) only when the task truly requires it, thus optimizing resource usage. AI computing resources.
Putting the plugin to the test: The onboarding case
Imagine you're new to the platform and need to set up a project from scratch with active invoicing and local authentication so your scripts work with a service identity. With the plugin installed, the agent doesn't launch blindly, but follows a logical process:
- Surrounding analysis: The wizard silently checks if you have the CLI installed and if any organizations or projects have already been created.
- Security review: Analyze the best practices in IAM to prevent, through oversight, private key filters in a Git commit.
- Roadmap: It proposes a detailed action plan and asks for your permission before touching any real resources in the cloud.
Step by step installation guide
Depending on the tool you use, the process varies slightly, but it always requires enabling the Developer Knowledge API using the command gcloud services enable developerknowledge.googleapis.com.
For Antigravity CLI users
This is the most direct option. You just need to run the installation command pointing to the path of the GitHub repository: agy plugin install https://github.com/google/skills/plugins/cloud/google-cloud-developer.
For Claude Code CLI
The process here is a bit more detailed. First, you add the Google Marketplace, and then you install the specific plugin. This is essential. generate an API key for the Developer Knowledge API and export it to your terminal using the environment variable DEVELOPERKNOWLEDGE_API_KEY, preferably by adding it to your file .bashrc o .zshrc so you don't have to do it every time.
For Codex CLI
As with Claude, the marketplace is used. The difference is that you need to update the configuration file. ~/.codex/config.toml To add the MCP server URL and the authentication header with your API key. Don't worry if you see a status of 'not logged in'It is normal behavior that does not affect operation.
Complementary tools and local environment
For all of this to perform at its best, it's vital to have a well-configured local environment. This involves installing the gcloud CLI, execute gcloud init for the default project and configure the Application Default Credentials (ADC) with gcloud auth application-default loginThis allows client libraries for Java, Python, or Go to authenticate themselves without complications.
This ecosystem transforms the way we develop in the cloud, moving from manual and error-prone management to a flow where AI has direct access to documentation and control tools, allowing the programmer to focus on business logic while the agent takes care of the technical complexity of the infrastructure.