Google Just Shipped an MCP Server. Here's Why That Matters for Your Business.
Google announced Workspace Intelligence at Cloud Next '26 today, and buried in the rollout was a line most coverage skipped past: a new Workspace MCP Server for outside AI apps and agents.
That sentence is the story.
For the last year, MCP (Model Context Protocol) has been a thing developers argued about on GitHub. Now it's shipping from Google, alongside Anthropic's existing commitment to the protocol, with Microsoft not far behind. If you run a small business and you've been waiting for AI to feel less like a novelty and more like plumbing, this is the week it started happening.
What was actually announced
Google used Cloud Next on April 22 to introduce three things that matter here:
- Workspace Intelligence — a semantic layer that maps your emails, chats, files, collaborators, and active projects into shared context for Gemini agents. The pitch is that AI stops treating Gmail, Drive, and Chat as separate apps and starts reasoning across them.
- Gemini auto browse in Chrome Enterprise — the agentic browsing feature Google launched for consumers in January, now coming to business accounts in the US. It can fill forms, compare options, and complete multi-step tasks across tabs.
- A Workspace MCP Server — the piece almost nobody led with. This lets external AI agents (not just Gemini) plug into Workspace through a standard protocol.
The first two are product news. The third is infrastructure.
Why the MCP part is the real headline
MCP is the protocol that lets an AI agent talk to your tools without someone writing a custom integration every time. Think of it the way USB-C replaced the drawer full of chargers. Before USB-C, every device had its own cable. Before MCP, every AI integration was bespoke.
When Google ships a Workspace MCP Server, it means any MCP-compatible agent — Claude, a custom agent your consultant built, an open-source tool from GitHub — can read and act on your Workspace data through a standard interface. You aren't locked into Gemini to get value from Gemini's context.
That's a quiet but significant shift in how AI rolls out in businesses. The question stops being "which AI vendor do I marry?" and starts being "which agents do I want touching which tools, and with what controls?"
What this means if you're a small business
Three practical things.
First, the "one AI to rule them all" pitch is dying. You aren't going to get a single subscription that does everything. You're going to have a handful of specialist agents that each do one job well, all plugged into your existing stack. Inbox triage, follow-ups, data sync, reporting — these are separate workflows with separate requirements.
Second, your tools matter more than your AI. If you're already on Google Workspace, you now have a supported path for agents to connect into Gmail, Drive, Calendar, and Chat. If you're on Microsoft 365, Microsoft has its own version of this coming. The AI layer is becoming portable. Your data and your workflows are the sticky bit.
Third, governance is no longer optional. Google's announcement specifically called out admin controls, agent governance tools, client-side encryption, and sovereign data controls. That isn't marketing fluff — it's a signal that enterprises are refusing to let agents loose on their data without a clear permission model. Small businesses should borrow that instinct. Know what each agent can see, what it can do, and who approves what.
The risk nobody is putting on the slide
Auto browse, by Google's own admission, is an experimental feature and users are responsible for what the agent does. That includes mistakes and unexpected purchases. Google also warned about prompt injection — where a malicious webpage or email hides instructions that trick the agent into leaking data or taking actions you didn't authorise.
This isn't a reason not to use agents. It is a reason to be thoughtful about which agents, doing what, with which permissions. A governance tier for "read-only research" is very different from a governance tier for "can send emails on my behalf" or "can make purchases."
If your AI rollout plan doesn't have a view on this, you don't have a plan. You have a demo.
So what should you actually do this week
Nothing urgent, if I'm honest. Workspace Intelligence is rolling out over the coming weeks and some of the companion features, including the MCP server, are still in preview.
What you can do is stop thinking about AI as a product you buy and start thinking about it as a layer of connective tissue between the tools you already use. Map the three or four workflows that genuinely eat your week — the inbox sorting, the follow-ups that slip, the weekly report nobody wants to write. Those are the places where agents plugged into your existing tools will earn their keep.
When you're ready to pick an agent, ask three questions: what does it connect to, what can it do without asking, and how do I see what it's done? If the vendor can't answer all three clearly, that's your answer.
The technology finally caught up to the promise. The hard part now is the judgement about how to use it.