Google, Apple, Samsung, and practically every other big tech firm are promising a future where AI will take over your phone and computer. However, as it currently stands, newfangled AI assistants aren’t as intelligent as their name suggests. Anthropic—the makers of the Claude chatbot—has a new tool that promises to give any assistant access to any app. If the future truly is based on “agentic AI,” then this model may be the way that Apple Intelligence or Gemini will collect the data from any of your devices’ apps.
The tool is called the Model Context Protocol, which Anthropic said on Monday it was making open source and free to use for any developer. This protocol will create an information channel between a data source, even app data, and an AI assistant. This is designed for use by AI agents like Anthropic’s Claude desktop app. Anthropic said this connection is “secure,” since—instead of exposing their app’s data—third-party developers can instead build a “MCP client,” AKA a separate AI application, to connect to their private servers.
While this ostensibly could be used by companies to allow any AI to access internal data more easily, it may also offer a new standard for how apps offer up their prized data to AI assistants. So far, AI apps like Google’s Gemini on both Android and iOS can access the tech giant’s own Workspace apps, so long as users give it permission. This means it can dig through your Gmail to surface some lost missives or turn any text reservations into an event on Google Calendar. Currently, Gemini doesn’t have access to third-party apps except through extensions.
Anthropic head of Claude relations, Alex Albert, claimed that developers would normally have to work directly with Google or any other AI assistant developer to create a custom connection between the AI and its data. MCP proclaims it solves that issue by enabling a free protocol that could work between the AI assistant and any app or data source.
Albert demoed Claude working directly with GitHub to make a new repository using MCP. He claimed that with MCP, it took “less than an hour” to build the AI-to-app integration. The company shared some repositories of sample MCP servers that it claims should help developers get started.
While the protocol is open source, Anthropic still promoted how the most recent Claude 3.5 Sonnet model was better for creating these data channels between servers and the AI. Otherwise, Anthropic said this tool will replace today’s “fragmented” integrations with something more “sustainable.” So far, Anthropic said coding software companies Zed, Replit, Codeium, and Sourcegraph have onboarded MCP for its AI integrations.
Apple has promised similar cross-app AI functionality for Apple Intelligence, eventually. That feature, alongside a rejuvenated Siri, could arrive in early spring next year. The Cupertino, California company’s famously secure walled garden may not have a place for MCP, but other, smaller AI agent developers may be eyeing this or any other similar protocol.
Either way, the major AI developers will be putting pressure on app makers to incorporate ways for AI assistants to access app data. While Apple claims its private cloud compute will keep all your personal data safe, we’ll have to see if any of this open-handed data transfer proves yet another privacy SNAFU.
{Categories} _Category: Platforms{/Categories}
{URL}https://gizmodo.com/anthropics-new-tool-shows-us-how-agentic-ai-will-take-over-phone-apps-in-2025-2000530087{/URL}
{Author}Kyle Barr{/Author}
{Image}https://gizmodo.com/app/uploads/2024/11/Anthropic-AI-apps-Gemini-Bing.jpg{/Image}
{Keywords}Artificial Intelligence,Anthropic,Artificial intelligence{/Keywords}
{Source}Platforms{/Source}
{Thumb}{/Thumb}
