Google Cloud has announced fully managed remote Model Context Protocol (MCP) servers, enhancing its existing API infrastructure to support MCP and providing a unified layer across all Google and Google Cloud services.
With the support for MCP servers, developers can point their AI agents or standard MCP clients, such as the Gemini CLI, to a globally consistent, enterprise-ready endpoint for Google and Google Cloud services. Initially, the company will incrementally roll out MCP support across all its services, starting with Google Maps, BigQuery, Google Compute Engine (GCE), and Google Kubernetes Engine (GKE).
The new support is seen as a strong endorsement for MCP (USB-C for AI) and a significant step toward wide-scale adoption beyond technically savvy users. However, the move has prompted discussion in a Reddit thread about whether “cloud MCP is solving a problem that running trusted MCPs already solves, and better,” especially regarding the potential benefits of running trusted MCP code locally with full edge compute access to improve latency. There is also a question among developers about whether remote MCPs are simply converting the protocol into a remote API similar to HTTP APIs.
To manage and secure MCP capabilities across services, Google introduces the new Cloud API Registry and Apigee API Hub, where developers can find trusted MCP tools from Google and their own organizations. Apigee can convert a standard enterprise API, such as a product catalog, into a discoverable MCP server, allowing organizations to expose their custom business logic to AI agents while maintaining existing governance and security.
Furthermore, these directories run in parallel to efforts such as the Agntcy project, a collaboration between Cisco, Google Cloud, Oracle, Red Hat, and Dell Technologies, which was donated to the Linux Foundation (LF) as well as the Agentic AI Foundation (AAIF) (which governs the protocol and anchors the LF’s neutral AI development infrastructure), highlights the industry-wide consensus on MCP.
Other hyperscalers are also deeply involved: Amazon Web Services (AWS) and Microsoft are Platinum members of the AAIF. Microsoft is integrating MCP directly into developer tools like Visual Studio Code and Copilot, while AWS provides extensive guidance and services, such as Amazon Bedrock AgentCore, to simplify the deployment of MCP servers on its cloud platform.
In a recent Medium blog post, Roman Irani:
The release of Google’s managed MCP services marks an interesting moment for developers working with Google services. Using the MCP Servers would go a long way toward removing a major friction point in agent development for agents integrating with various Google Services.
Lastly, the MCP servers are currently available in public preview. A demo with MCP for the supported Google services is available on GitHub.
