AI agents are often promoted as versatile solutions for planning trips, answering business questions, and tackling a wide range of problems. However, integrating these agents with external tools and data beyond their chat interfaces has remained a complex challenge. Developers typically need to assemble various connectors and maintain them, an approach that is not only fragile and difficult to scale but also introduces significant governance issues.

Google is now aiming to address this by introducing its own fully managed, remote MCP servers. These servers are designed to make Google and Cloud services—such as Maps and BigQuery—more accessible for AI agents to connect with. This initiative follows the recent launch of Google’s latest Gemini 3 model, as the company seeks to combine advanced reasoning capabilities with more reliable connections to real-world tools and data.

Instead of spending one to two weeks configuring connectors, developers can now simply paste a URL to a managed endpoint, according to Giannini. At launch, Google is offering MCP servers for Maps, BigQuery, Compute Engine, and Kubernetes Engine. In practical terms, this could enable an analytics assistant to query BigQuery directly or an operations agent to interact with infrastructure services.

For example, with Maps, Giannini noted that without MCP, developers would depend on a model’s built-in knowledge. “But by giving your agent […] a tool like the Google Maps MCP server, then it gets grounded on actual, up-to-date location information for places or trips planning,” he explained.

While these MCP servers will eventually be available across all Google tools, they are initially launching in public preview. This means they are not yet fully covered by Google Cloud terms of service. However, they are being offered at no additional cost to enterprise customers who already pay for Google services. “We expect to bring them to general availability very soon in the new year,” Giannini said, adding that he anticipates more MCP servers will be introduced on a weekly basis.

MCP, which stands for Model Context Protocol, was developed by Anthropic approximately a year ago as an open-source standard for connecting AI systems with data and tools. The protocol has gained widespread adoption across the agent tooling ecosystem. Earlier this week, Anthropic donated MCP to a new Linux Foundation fund dedicated to open-sourcing and standardizing AI agent infrastructure.

“The beauty of MCP is that, because it’s a standard, if Google provides a server, it can connect to any client,” Giannini remarked. “I’m looking forward to seeing how many more clients will emerge.”

MCP clients can be thought of as the AI applications on the other end that communicate with MCP servers and utilize the tools they provide. For Google, this includes Gemini CLI and AI Studio. Giannini mentioned that he has also tested it with Anthropic’s Claude and OpenAI’s ChatGPT as clients, and “they just work.”

Google emphasizes that this effort extends beyond merely connecting agents to its services. A broader enterprise strategy involves Apigee, its API management product, which many companies already use to issue API keys, set quotas, and monitor traffic. Giannini explained that Apigee can effectively “translate” a standard API into an MCP server, transforming endpoints like a product catalog API into tools an agent can discover and use—all while maintaining existing security and governance controls. In essence, the same API safeguards that companies apply to human-built applications can now also govern AI agents.

Google’s new MCP servers are secured by a permission mechanism called Google Cloud IAM, which explicitly controls what an agent can do with a server. They are also protected by Google Cloud Model Armor, described by Giannini as a firewall dedicated to agentic workloads that defends against advanced threats such as prompt injection and data exfiltration. Administrators can additionally rely on audit logging for enhanced observability.

Looking ahead, Google plans to expand MCP support beyond the initial set of servers. Over the next few months, the company will roll out support for services across areas including storage, databases, logging and monitoring, and security. “We built the plumbing so that developers don’t have to,” Giannini concluded.