In a nutshell: Google AI Edge Gallery now supports the Model Context Protocol (MCP), enables local notifications, and provides persistent chat history. Developers can use it to build and test intelligent, connected AI agents directly on mobile devices.
Google is extending its AI Edge Gallery with the Model Context Protocol (MCP), local notifications, and persistent chat history. Developers can now create and test connected, autonomous AI agents directly on mobile devices.
The Google AI Edge Gallery, an on-device showcase app for Android and iOS, is receiving significant new features to support advanced AI applications. Users and developers can now run advanced, agent-based workflows directly on mobile devices with Gemma and other open-source models.
The centerpiece of the update is support for the open Model Context Protocol (MCP). This allows the on-device model to interact with the world outside the app sandbox in a standardized way. Developers can register valid MCP URLs in the app, which imports tool definitions and resource schemas directly into the model’s system prompt. The reasoning and decision-making processes take place entirely on the smartphone. When users ask a question, Gemma 4 automatically determines which tool is needed and executes the call locally.
This architecture enables mobile devices to coordinate complex tasks across various data sources and functional tools. Practical use cases demonstrate the potential: with Google Workspace MCP, mobile agents can query calendars or search inboxes for invoices and ticket information. The Google Maps MCP allows for natural language queries about nearby places or travel times. A Web Fetch MCP integration allows agents to retrieve and analyze content from URLs to stay informed about news or documentation in real time.
The experimental feature is initially available in the Android app, with an iOS update coming soon. Developers are advised to keep tool descriptions brief, as on-device models work with smaller context windows than server-side models.