Gemini

Gemini Spark Comes to Mac: Google's Agent Takes the Desktop

Google's 24/7 agentic assistant Gemini Spark lands on the Mac in beta for AI Ultra subscribers, adding Tasks and Keep, five third-party apps, and custom MCP support.

Gemini Spark Comes to Mac: Google's Agent Takes the Desktop — article cover

On June 30, 2026, Google used its official blog to announce that Gemini Spark, its 24/7 agentic assistant, is heading to macOS: a Mac build has entered beta, folded into the existing Gemini desktop app, and initially open to Google AI Ultra subscribers in the US aged 18 and up, via gemini.google/mac. The same announcement brought a batch of new app integrations and custom MCP support, with everything rolling out from that day.

Spark is the assistant Google launched in late May, built to handle your digital life around the clock in the background. Bringing it to the Mac moves the battlefield from the cloud onto the desktop, straight into territory held by Claude Desktop, Microsoft’s Copilot, and OpenClaw.

What Spark Can Do on the Desktop

The desktop pitch is simple: it can touch your files. Spark can organize documents on the computer, for example sorting scattered PDFs in the Downloads folder into the right folders, and it can use local files as source material for new Google Workspace documents or spreadsheets. Google’s demo scenario turns invoices into a budgeting worksheet, with scheduled updates flowing through the Workspace integration so the spreadsheet keeps itself current. On permissions, the design is deliberately conservative: Spark only accesses files the user explicitly allows, which matters for a service that reads anything as sensitive as invoices.

The next step Google teased is remote assignment: soon users will be able to dispatch multi-step tasks from their phone, such as asking the desktop agent to locate a sales report, extract the revenue figure, and email it to someone, while the Mac grinds through the work at home. That turns the desktop build from a convenience feature into a genuine endpoint in a cross-device agent system.

New Connections: Tasks, Keep, and Five Third-Party Apps

Integrations are the other headline of this update. On the Google side, Tasks and Keep are finally connected; Keep’s absence was the single biggest complaint in TechCrunch’s hands-on test of Spark in May, and now Keep notes can be turned into Tasks action items automatically. On the third-party side, five services joined at once: Canva, Dropbox, Instacart, OpenTable, and Zillow Rentals, each mapped to concrete errands such as designing a flyer, sharing files, ordering groceries, booking a restaurant, or scheduling an apartment tour. These connections are rolling out on web and mobile over the following week, with macOS support following in the coming weeks.

Live Monitoring and Custom MCP

Spark is also strengthening real-time tracking: blogs, news, social media, finance, shopping, weather, sports, and email are all fair game. Soccer highlights can arrive when the match ends, and a financial report can fire the moment a stock crosses a threshold you set, so you stop refreshing pages yourself.

The more strategically interesting piece is custom MCP support: users can use the Model Context Protocol to connect their favorite apps directly to Spark and assemble a more tailored assistant. For power users, that turns Spark from “take what Google gives you” into a platform they can extend themselves.

The Paywall and the Competition

The limits are just as clear. The Mac build is a beta, available only to AI Ultra subscribers in the US, and Ultra is the most expensive tier in Google’s subscription lineup. Tying the most aggressive agentic capability to the flagship plan does two things at once: it contains risk and cost, and it hands Ultra subscribers a harder reason to keep paying.

On the competitive map, desktop agents are becoming the main arena of consumer AI. Claude Desktop, Copilot, and OpenClaw have all staked claims, and each brings a different asset: Anthropic has the developer mindshare, Microsoft has Windows, and the open-source contenders have price. Google’s edge is that it simultaneously owns the browser, Workspace, Android, and an enormous account graph to wire together, which no rival can fully replicate. The bet is that an assistant which can see your files, your calendar, and your inbox at once will beat a smarter model confined to a chat window.

There is also a defensive reading. If agentic assistants become the primary way people get computing done, whoever owns the agent owns the user relationship, and Google has more of that relationship to lose than almost anyone. Google has promised more Spark news over the summer; the Mac build is only the first move in that game.

Sources

AI-assisted summary compiled from the sources above, reviewed by a human before publishing.

SHAREXEMAIL