Google Cloud plugin gives Antigravity agents live documentation
Google’s Antigravity Marketplace plugin grounds agents in official cloud documentation, validates gcloud commands, limits context and requires consent before billable APIs or infrastructure changes.

Why it matters
- November 17, 2026: Google made the Google Cloud Developer plugin available through the Antigravity Marketplace.
- One command: command-line installation registers the skills and connects the documentation server immediately.
- No API keys, OAuth token exchange or configuration-file edits are required before connecting to official documentation.
- User consent is required before agents enable billable APIs or modify cloud resources.
- The gcloud skill validates Google Cloud CLI syntax and applies execution checks.
On November 17, 2026, Google made its Google Cloud Developer plugin available through the Antigravity Marketplace. Agents using the plugin must obtain user consent before enabling billable APIs or modifying cloud resources.
Google said the plugin “connects to official Google developer documentation immediately.” The plugin links Antigravity agents to Google Cloud documentation and gives coding agents cloud-specific guidance when they build or manage services on Google Cloud.
The release targets a weakness Google describes in coding assistants that rely on static model training. Cloud platforms change frequently, and coding assistants can return code that no longer matches current client libraries or platform parameters, according to Google.
Google says the plugin is intended to reduce three common errors: outdated SDK syntax, invalid command flags and tasks attempted without the necessary Identity and Access Management permissions. Its controls also address actions that could create costs or alter infrastructure.
Google’s design starts with three operational risks: a client call that no longer matches the SDK, a command flag the platform does not accept and a task that exceeds available Identity and Access Management permissions. The package puts current references, command validation and approval gates near execution rather than leaving those checks to model memory.
What does the plugin add to Antigravity?
Google says the package combines several bundled skills with a pre-authorized Developer Knowledge Model Context Protocol server. The server supplies official documentation, and users need no API keys, OAuth token exchange or configuration-file edits before connecting an agent to that documentation.
The authentication arrangement differs from many Model Context Protocol implementations, where developers handle authentication directly. In the Antigravity integration, Marketplace access pre-authorizes the Developer Knowledge server, while Google manages authentication in the background.
Google said the plugin grounds answers in live documentation delivered over HTTP. It also inspects local command-line settings so an agent can map an execution plan before running commands.
Which skills ship with the package?
Google listed these bundled components:
finding-google-skillsdiscovers and loads additional task-specific skills from Google’s public repository when a task needs them.retrieving-developer-knowledgesupplies answers and code samples grounded in official documentation for Google Cloud, Firebase, Android, Web and Gemini.- The
gcloudskill validates command syntax and applies execution checks to Google Cloud CLI commands. google-cloud-recipe-authdirects agents toward service identities and Application Default Credentials. It also blocks the download of service account JSON keys.google-cloud-recipe-onboardingreviews project state, account setup and billing before services are activated.
In Google’s design, each component covers a different stage of an agent’s work. The retrieval skill supplies current documentation, the gcloud skill checks command form, the auth recipe constrains credential handling and the onboarding recipe reviews whether a project is ready for service activation.
How does Google keep an agent’s context focused?
Google added a routing rule named google-cloud-discovery.md. It works with finding-google-skills to pull specialist guidance only when a prompt requires it, rather than loading a broad set of niche skills in advance, Google said.
The rule limits how much context an agent holds at one time while preserving access to deeper instructions for specific tasks. Google describes this as a way to keep the agent focused on relevant information rather than maintain a fixed, broader set of tools.
What stops an agent from spending money or changing infrastructure?
Google said every agent using the plugin must obtain user consent before it enables a billable API or modifies a cloud resource. The rule gives users a control point for two classes of action with direct financial or operational consequences: paid API use and infrastructure changes.
The gcloud skill adds a command-level check by validating syntax and applying execution checks before commands proceed. The local-settings inspection gives the agent a view of the command environment so it can map the execution plan first, according to Google.
google-cloud-recipe-auth narrows the credential path as well. It directs agents toward service identities and Application Default Credentials while blocking the download of service account JSON keys, Google said.
google-cloud-recipe-onboarding checks project state, account setup and billing before Google Cloud services are activated. The recipe turns those three preconditions into a review before activation.
Taken together, the controls cover retrieval, syntax, credentials, account readiness and authorization. They also retain a user decision point for paid API activation and resource modification, which Google says can otherwise create costs or alter infrastructure.
How do developers install the plugin?
Developers can install the plugin from the Antigravity Marketplace interface or through the command line, Google said. The command-line route uses a single command. After that command, Google registers the skills and connects the documentation server immediately.
Google illustrated the plugin with a comparison query covering Google Kubernetes Engine and Cloud Run. With the plugin installed, the agent answers from current documentation rather than returning a model response that may be outdated, Google said.
Why does distribution through a marketplace matter?
The release shows how cloud providers increasingly seek to shape the behavior of third-party coding agents. Google supplies its own documentation links, execution checks and policy controls at the point of installation, where developers choose the tools an agent can use.
For developers, Marketplace installation removes the initial OAuth token exchange and configuration-file work for the documentation server. For Google, the same distribution point gives the cloud provider a way to keep changing guidance, credential practices and approval requirements in front of agents built around static training data.
That arrangement matters because Google’s own documentation changes, while the command settings that affect execution remain local to each developer. Google’s package addresses those two sources of variation with live documentation and local-settings inspection, then adds credential, billing and consent checks around the resulting actions.
The next test is whether the combination remains reliable as Google Cloud changes. Agents need current references, valid commands, usable credentials and explicit approval before paid or infrastructure-changing actions. Google’s package places each requirement at a different stage of the workflow, from Marketplace authentication through command checks and pre-action consent.
Source: itbrief.com.au
More from Rebecca Stone
Show full bio
Correspondent covering companies, deals and strategy in artificial intelligence and machine learning at AI In Context.
728 articles
Related articles
- Google Cloud launches IAM Helper to let agents manage access
- Google Launches Antigravity, an Agent-First IDE Built for Gemini 3
- Google opens Android Studio to 40-plus coding agents
- Google Cloud Opens 'Gemini Factory' to Speed Partner AI Agent Builds
- Google Cloud ships new security tools for autonomous AI agents