Gemini Spark Comes to Mac With File Access, Real-Time Tracking, and Custom MCP

Google added Gemini Spark to its Mac desktop app on July 1, giving the agentic AI assistant its first desktop home outside mobile. Gemini Spark on Mac is currently limited to Google AI Ultra subscribers in the U.S. - the same tier required when Spark launched last month. Alongside the Mac rollout, Google pushed a batch of updates including real-time topic tracking, new app integrations, and custom MCP connector support.

Files on Mac, With Remote Tasks Coming Later

Spark on Mac can access and organize files on the local machine, and use them as source material for Google Workspace documents or spreadsheets. Google's example: invoices in, budgeting worksheet out. Remote tasks - where the mobile app calls up the desktop agent to pull a file from the Mac - are listed as coming soon but not live at launch.

Mac access puts Spark into more direct competition with desktop AI agents like OpenClaw, Claude Desktop, and Microsoft Copilot, all of which already ship with local file access. OpenClaw routes nothing through third-party servers - data stays on the user's own hardware - while Spark moves through Google's infrastructure. For teams with strict data requirements, that difference is the whole decision.

Keep and Tasks Are In, Plus Five Third-Party Apps

Google addressed a gap flagged in early Spark reviews by adding Google Tasks and Google Keep integration, both absent from the original launch. Short notes belong in Keep, not Docs. Alongside those two, Spark now connects to Canva, Dropbox, Instacart, OpenTable, and Zillow Rentals, which means Spark can now design flyers, order weekly groceries, reserve restaurant tables, or book apartment tours through those platforms directly.

Real-Time Tracking and Custom MCP Support

Spark can now monitor topics and react to live changes - sports scores, stock movements, weather, and breaking news - rather than relying on a user prompting it for updates. Standard across many AI assistants now. Google is also rolling out custom Model Context Protocol connections, letting users wire their own apps directly into Spark and build a more personalized agent. MCP has become the standard connector layer across Claude, Cursor, and most other AI platforms over the past several months; Spark's support brings it to Google's agent stack.

Google has not announced a timeline for expanding Spark beyond AI Ultra subscribers or opening access to non-U.S. markets.