Google Cloud Next 2026: MCP Becomes a First-Class Cloud Primitive

Google Cloud Next 2026 wrapped up today in Las Vegas. The headline: MCP is no longer an experiment at Google. It is a production-grade integration layer across Workspace, databases, Vertex AI (now Gemini Enterprise Agent Platform), and developer tooling.

This is the most aggressive MCP adoption we have seen from any hyperscaler. Here is what shipped.

Workspace MCP Server (Preview)

Google announced a Workspace MCP Server that exposes Drive, Gmail, Calendar, and Chat as MCP resources and tools. An agent connected to this server can search and synthesize Drive documents, draft Gmail messages, create calendar events, and post to Chat spaces. All through standard MCP tool calls.

This matters because Workspace has over 3 billion users. Exposing it through MCP means any compliant agent, not just Gemini, can interact with Workspace data. Claude, ChatGPT, and open-source agents all speak the same protocol. Google is betting that openness here drives more Workspace adoption, not less.

The server is in preview now with GA expected later this year.

Managed Remote MCP Servers for Databases

Google is now hosting MCP infrastructure for database connectivity. Instead of running your own MCP server to bridge an agent to Cloud SQL, Spanner, or AlloyDB, Google manages the server for you. You configure access, Google handles the runtime, auth, and scaling.

This removes the biggest friction point in database MCP adoption: operations. Running a persistent MCP server with proper connection pooling, auth rotation, and monitoring is real work. Google is absorbing that cost.

The managed servers support read and write operations with configurable permissions. Admins can restrict an agent to read-only access on specific tables or schemas.

Database Center MCP: Now GA

Database Center MCP moved from preview to general availability this week. You can now connect from Gemini CLI, ChatGPT, or Claude directly to your Google Cloud databases through a centralized MCP endpoint.

Database Center acts as a catalog layer. It knows about your Cloud SQL instances, Spanner databases, BigQuery datasets, and AlloyDB clusters. Agents query the catalog to discover what data is available, then interact with specific databases through the same MCP connection.

GA means SLA-backed availability and enterprise support. Production workloads can depend on it.

BYO-MCP for Gemini Enterprise

Gemini Enterprise (the platform formerly known as Vertex AI) now supports bring-your-own MCP servers. Enterprise admins can register custom or third-party MCP servers and make them available to Gemini-powered agents across their organization.

This is the connector model that enterprise buyers have been asking for. You build an MCP server that wraps your internal APIs, register it with Gemini Enterprise, and every agent in your org can discover and use those tools. IT controls which servers are approved. Developers consume them without managing infrastructure.

Third-party ISVs can also publish MCP servers that enterprise admins opt into. Expect Salesforce, ServiceNow, and similar vendors to ship certified MCP servers for this marketplace.

Agents CLI

Google released Agents CLI, a unified command-line tool for the agent development lifecycle. It handles project scaffolding, local testing, deployment, and monitoring for agents built on the Gemini Enterprise Agent Platform.

The interesting part: Agents CLI is not locked to Gemini. It works with Gemini CLI, Claude Code, and Cursor. You can develop an agent using Claude Code locally, test it with Agents CLI, and deploy it to Google’s managed runtime. The CLI manages MCP server connections, environment variables, and agent configuration across all three development environments.

This is a pragmatic move. Google knows developers use multiple AI coding tools. Meeting them where they are reduces friction to deploying on Google’s platform.

Data Agent Kit

Data Agent Kit is a new set of portable MCP tools and plugins for working with structured data. It ships as extensions for VS Code and Gemini CLI.

The kit includes MCP tools for common data tasks: schema exploration, query generation, data profiling, and result visualization. These tools connect to any database that has a registered MCP server, whether managed by Google or self-hosted.

Think of it as a starter pack for building data agents. Instead of writing custom tool definitions for every database operation, you install the Data Agent Kit and get a working set of tools immediately.

Vertex AI Rebranded to Gemini Enterprise Agent Platform

The rebrand signals where Google sees the product heading. Vertex AI started as an ML platform. It evolved into an AI platform. Now it is explicitly an agent platform.

The new name, Gemini Enterprise Agent Platform, puts agents at the center. MCP is the integration protocol. A2A (Agent-to-Agent) is the communication protocol. Gemini models are the default runtime. But the platform supports external models and external MCP servers, making it more of an orchestration layer than a walled garden.

The Bigger Picture

The MCP ecosystem has reached 17,000+ indexed servers with 97 million monthly SDK downloads. Google’s announcements this week matter because they validate MCP at the infrastructure level. This is not a plugin or an add-on. Google is building MCP into the control plane of its cloud.

Google also launched the A2A (Agent-to-Agent) protocol at the same event. A2A handles cross-platform agent communication, letting agents built on different frameworks and by different vendors collaborate on tasks. MCP handles tool integration. A2A handles agent coordination. Together they form a two-protocol stack for multi-agent systems.

This was the biggest week for MCP adoption by a hyperscaler. AWS and Azure have MCP support in various services, but neither has made it this central to their platform narrative. Google is staking a position: if MCP is going to be the universal agent integration layer, Google wants to be the best place to run MCP-connected agents.

FAQ

Can I use non-Google AI models with Google’s MCP servers?

Yes. Google’s MCP servers follow the open MCP specification. Any MCP-compliant client can connect, including Claude, ChatGPT, and open-source models. The Workspace MCP Server, Database Center MCP, and managed database servers all work with external clients.

What is the difference between A2A and MCP?

MCP (Model Context Protocol) connects AI models to tools and data sources. A2A (Agent-to-Agent) connects agents to each other. MCP is about what an agent can do. A2A is about how agents collaborate. They are complementary protocols, not competing ones.

Do I need to migrate from Vertex AI to use these features?

No migration is needed. Gemini Enterprise Agent Platform is a rebrand of Vertex AI, not a new product. Existing Vertex AI projects, APIs, and configurations continue to work. The new MCP features are additive. You enable them in your existing environment.