Google is rolling out a new infrastructure layer aimed at solving one of the biggest challenges facing AI agents today: reliably connecting them to real-world tools, services, and enterprise data. The company has launched fully managed, remote Model Context Protocol (MCP) servers designed to let AI agents seamlessly plug into Google and Google Cloud services without complex setup or ongoing maintenance.
AI agents are increasingly promoted as digital assistants capable of planning travel, analyzing data, and managing operations. However, making those agents interact with external tools has often required fragile, custom-built connectors that are difficult to scale and govern. Google’s new approach aims to replace that patchwork with standardized, managed endpoints.
The launch closely follows Google’s release of its latest Gemini 3 model and reflects a broader push to pair stronger reasoning capabilities with dependable access to live data and services. According to Google Cloud, the goal is to make its ecosystem “agent-ready by design,” reducing the technical barriers developers face when deploying AI-powered workflows.
Instead of spending days or weeks configuring and maintaining integrations, developers can now connect agents to Google services by pointing them to a managed MCP endpoint. At launch, Google is offering MCP servers for key services including Google Maps, BigQuery, Compute Engine, and Kubernetes Engine. This enables use cases such as analytics agents querying live datasets in BigQuery or operations agents interacting directly with cloud infrastructure.
For location-based tasks, the impact is especially notable. Without MCP, agents rely largely on their internal training data. With a Google Maps MCP server, agents can access up-to-date location information, enabling more accurate trip planning and local recommendations grounded in real-time data.
The MCP servers are currently available in public preview and are not yet fully covered under standard Google Cloud terms. However, they come at no additional cost for enterprise customers already using Google’s cloud services. Google says general availability is expected early next year, with new MCP servers rolling out on a weekly basis.
MCP itself is an open-source standard originally developed by Anthropic to connect AI systems with tools and data. Its growing adoption has positioned it as a common language for AI agents. Because MCP is standardized, Google’s servers can work with a wide range of AI clients, including Gemini tools as well as third-party systems such as Anthropic’s Claude and OpenAI’s ChatGPT.
Beyond its own services, Google sees a major enterprise opportunity in Apigee, its API management platform. Apigee can effectively convert existing APIs into MCP-compatible servers, allowing organizations to expose internal systems to AI agents while maintaining the same security, quotas, and governance controls used for traditional applications.
Security and oversight are central to the design. Google’s MCP servers are protected by Cloud IAM permissions, Model Armor defenses against threats such as prompt injection, and detailed audit logging. In the coming months, Google plans to expand MCP support across storage, databases, monitoring, and security services—further embedding AI agents into the core of enterprise infrastructure.
