Artificial intelligence takes a new step toward real autonomy. Google Cloud has just formalized support for the Model Context Protocol (MCP) for a wide range of its services, from Google Maps to BigQuery and Google Kubernetes Engine (GKE). Often compared to a “USB-C for AI,” this standardized protocol now allows developers to easily connect their intelligent agents to Google's infrastructure without having to manage complex local servers. In short: a breakthrough that promises to unlock the potential of autonomous agents in business.
Key takeaways:
- Mass standardization: Google integrates the MCP protocol into its existing APIs, offering unified, fully managed access to its services without requiring fragile local servers.
- Connection to real-world data: Agents can now natively interact with Google Maps (geospatial data), BigQuery (enterprise data analytics) and cloud infrastructure (GCE, GKE) to perform complex tasks.
- Security and governance: The integration includes robust safeguards via Google Cloud IAM and Model Armor to prevent threats such as indirect prompt injections.
- Enterprise extension: Through Apigee, companies can expose their internal APIs as discoverable “tools” for AI agents.
The end of "jury-rigged" integrations
Until now, connecting an AI agent (like Gemini or Claude) to external data or tools often required fairly heavy engineering. Developers had to build, host and maintain their own intermediary servers to bridge the model and the tool. It was costly and technically fragile.
With this December 11, 2025 announcement, Google changes the game. Instead of asking developers to create these bridges, Google has updated its API infrastructure so that it natively "speaks" the MCP language. Concretely, this means a developer can point their AI agent (whether running on Gemini 3 or another MCP client) directly at a endpoint reliable, secure Google
As Michael Bachman and Anna Berenberg of Google Cloud point out, for an AI to truly be an "agent," it must not only be intelligent; it must be able to reliably manipulate tools and data.
The first compatible services: from maps to the cloud
Google is rolling out this support in waves, but the first integrated services already cover critical needs:
1. Grounding in the real world with Google Maps
Thanks to Maps Grounding Lite, AI agents gain access to fresh geospatial data. No more hallucinations about travel times or whether places are closed. An agent can now answer complex logistical questions precisely—such as "What is the distance between the nearest park and this rental?"—or plan routes while accounting for the weather, all via a standardized connection.
2. Data analysis with BigQuery
This is a major asset for businesses. The MCP server for BigQuery allows agents to understand database structures (schemas) and execute SQL queries directly. The advantage is twofold: it avoids copying sensitive data into the model's context window (which is risky and costly) and it enables the agent to use BigQuery's advanced features, such as forecasts (forecasting).
3. Infrastructure management (GCE and GKE)
AI becomes an autonomous system administrator.
- On Google Compute Engine (GCE), agents can provision or resize virtual machines on demand.
- On Google Kubernetes Engine (GKE), the MCP interface structures interactions with clusters. The agent no longer merely reads raw text output from the console; it interacts properly with the Kubernetes API to diagnose failures or optimize costs, under human supervision or autonomously.
Apigee: the bridge to your own data
The announcement goes beyond Google services alone. By integrating MCP with Apigee, Google enables companies to transform their own tech stack into tools for AI.
Your internal APIs—the ones that manage your inventory, customers, or specific business processes—can be exposed and governed as MCP tools. They become "discoverable" by agents, allowing the creation of hybrid workflows that combine Google's computing power with your proprietary business logic.
Security and the future of the ecosystem
One of the main barriers to adopting autonomous agents remains security. Google addresses this concern by wrapping these MCP servers in its usual security layers. Access is managed by Cloud IAM (identity management), observability is ensured by audit logs (audit logging), and Model Armor protects against new threats specific to LLMs, such as injections via third-party data.
The goal is clear: to build an ecosystem where models, whether Gemini 3 or others, can thrive. Besides, Google is already announcing what comes next. In the coming months, MCP support will expand to Cloud Run, Cloud Storage, Spanner, Cloud SQL and even security tools (SecOps).
As founding member of theAgentic AI Foundation alongside AnthropicGoogle is confirming here its intention not only to provide the smartest models but, above all, the most robust infrastructure to make them work.
The article “Google adopts Anthropic's MCP standard to turn its services into tools for AI agents” was published on the site Abondance.
