AI agents are being sold as a solution for planning trips, answering business questions, and solving all kinds of problems, but getting them to interact with tools and data outside of a chat interface has been difficult. Developers have to piece together various connectors to keep them running, which is a brittle approach that is difficult to scale and poses governance issues.
Google claims it is trying to solve this problem by launching its own fully managed remote MCP server, which will make it easier for agents to connect its Google and cloud services, such as Maps and BigQuery.
The move follows the launch of Google’s latest model, Gemini 3, in which the company aims to combine more powerful inference with more reliable connections to real-world tools and data.
“We are designing for Google Agent,” Steren Giannini, director of product management at Google Cloud, told TechCrunch.
Giannini said that instead of spending a week or two setting up a connector, developers can now essentially paste a URL to a managed endpoint.
At launch, Google will start with MCP servers for Maps, BigQuery, Compute Engine, and Kubernetes Engine. In practice, this might look like an analytics assistant that queries BigQuery directly, or an operational agent that interacts with infrastructure services.
In the case of maps, Giannini said that without MCP, developers would rely on knowledge built into the model. “But if you give agents (…) tools like Google Maps MCP Server, they can plan locations and trips based on real, up-to-date location information,” he added.
tech crunch event
san francisco
|
October 13-15, 2026
MCP Server will eventually be available in all Google tools, but it will initially launch in public preview, so it’s not yet fully covered by Google Cloud’s Terms of Service. However, it is available at no additional charge to business customers who already pay for Google services.
“We plan to make it generally available as soon as the new year,” Giannini said, adding that he expects more MCP servers to be rolled out a little each week.
MCP stands for Model Context Protocol and was developed by Anthropic about a year ago as an open source standard for connecting AI systems with data and tools. The protocol has been widely adopted across the agent tools world, and earlier this week Anthropic donated MCP to a new Linux Foundation fund dedicated to open sourcing and standardizing AI agent infrastructure.
“The benefit of MCP is that it’s a standard, so once Google provides the server, any client can connect to it,” Giannini said. “I’m looking forward to seeing how many customers we’ll have in the future.”
You can think of the MCP client as an AI app on the other end of the wire that communicates with the MCP server and calls the tools provided by the MCP server. For Google, this includes Gemini CLI and AI Studio. Giannini said he tried Anthropic’s Claude and OpenAI’s ChatGPT as clients and said they “worked great.”
Google claims this is more than just connecting agents to its services. A larger enterprise effort is Apigee, an API management product. This is already used by many companies to issue API keys, set up quotas, and monitor traffic.
Giannini said Apigee essentially “translates” standard APIs into MCP servers, turning endpoints like the Product Catalog API into tools that agents can discover and use, and layer existing security and governance controls on top.
In other words, the same API guardrails that companies use for human-written apps will now apply to AI agents.
Google’s new MCP servers are secured by an authorization mechanism called Google Cloud IAM, which explicitly protects what agents can do on that server. They are also protected by Google Cloud Model Armor. Giannini describes it as a firewall specifically for agent workloads that protects against advanced agent threats such as prompt injection and data leakage. Administrators can also utilize audit logs for increased observability.
Google plans to expand MCP support beyond the initial set of servers. In the coming months, the company plans to roll out support for services across areas such as storage, databases, logging and monitoring, and security.
“We did the plumbing so the developer didn’t have to do any plumbing,” Giannini said.
