Google Cloud releases dev tools

Google Cloud has put its own documentation into a format that artificial intelligence tools can read directly, releasing a Developer Knowledge Application Programming Interface and a Model Context Protocol server in public preview. (developers.googleblog.com) Google said the tools give agents and coding assistants a “canonical, machine-readable gateway” to official documentation for products including Firebase, Google Cloud and Android. The company said developers can start using both tools now with an Application Programming Interface key in a Google Cloud project. (developers.googleblog.com) A Model Context Protocol server is a standardized bridge that lets an artificial intelligence application discover tools and call outside services over a common interface. Google Cloud’s documentation says remote servers run on Google infrastructure and expose capabilities over Hypertext Transfer Protocol endpoints. (docs.cloud.google.com) Google has been widening that Model Context Protocol push across its cloud products. In the past month, the company added managed servers for AlloyDB, Spanner, Cloud SQL, Firestore, Bigtable and a Developer Knowledge server aimed at connecting integrated development environments to Google documentation. (cloud.google.com) The immediate problem Google is trying to solve is stale or hallucinated answers from coding assistants. Its developer blog said builders need assistants that know the latest Firebase features, recent Android Application Programming Interface changes and current Google Cloud best practices. (developers.googleblog.com) Google paired that documentation push with a fresh pitch for Gemma 4, its open model family announced on April 2. Google DeepMind said Gemma 4 is built for reasoning and agent workflows, and described it as an open model line with more than 400 million cumulative downloads across the broader Gemma family and more than 100,000 community variants. (blog.google) Google said Gemma 4 is designed to run efficiently enough for single-graphics-processing-unit deployments, a lower hardware bar than many frontier models require. Google Cloud said customers can deploy Gemma 4 on Vertex Artificial Intelligence, Cloud Run, Google Kubernetes Engine and Sovereign Cloud. (blog.google, cloud.google.com) Google is also adding tool-use features to smaller open models. In January, the company introduced FunctionGemma, saying the model was built for native function calling so on-device agents could trigger actions such as reminders or system settings instead of only generating text. (blog.google) The combined message is that Google wants developers to have both cleaner inputs and cheaper models: official docs for answers, and lighter open weights for deployment. That leaves Google competing not just on model quality, but on the plumbing around how agents fetch facts and take actions. (developers.googleblog.com, blog.google, cloud.google.com)