Google Opens Gemini and Search to Outside Apps With Connected Apps Platform

Google launched a public preview of its Connected Apps Platform, letting U.S. developers build a single integration that runs inside both Google Search and the Gemini app. Developers connect via a cloud-hosted Model Context Protocol server or Android MCP using AppFunctions APIs, then register, test and submit apps through the Google Connected Console, choosing whether to appear in Search, Gemini or both, and tracking traffic, latency and error rates afterward.

More than 50 connected apps are already live in Gemini, including Adobe, Experian, Localiza and Peloton, with those heading to Search in coming weeks; Calm is a new partner. Access will expand in phases, and Google did not give a broader launch date. Google is inviting outside developers to plug their services directly into its two biggest consumer surfaces, launching a preview of a platform that lets third-party apps run inside both Google Search and the Gemini chatbot.

The company said the Connected Apps Platform, announced Wednesday, gives developers a single build that can reach users across both products. It is available in public beta to U.S. developers, who can join a waitlist starting today. Under the program, developers connect through a cloud-hosted Model Context Protocol server or an Android MCP using AppFunctions APIs.

They then register, test and submit their app through the Google Connected Console, choosing before review whether it will appear in Search, in Gemini, or in both. Once an app is live, its maker can monitor traffic, latency and error rates and retain control over branding, including verified logos and authenticated accounts.

Google said access will be opened in phases, with additional developers, markets and Google products added over the coming months. The move extends an effort Google began earlier this year, when it introduced connected apps that let people link third-party services to its AI tools and act on them directly -- designing a flyer through Canva, for instance, or assembling a playlist in YouTube Music.

Google says more than 50 such integrations are already live in Gemini, spanning productivity, event ticketing, health and fitness. Partners include Adobe, Experian, Localiza and Peloton, and those apps are slated to reach Search in the coming weeks. Search currently carries a smaller set, among them Instacart, Canva, YouTube Music and Yahoo Fantasy Football.

Google also named Calm as a new partner, saying the meditation app will launch on both Gemini and Search. The strategy carries competitive weight. Rivals including OpenAI's ChatGPT and Meta's Muse have pursued similar integrations, betting that consumers will increasingly reach their favorite services through an AI assistant rather than by opening individual apps.

For Google, a broad catalog of partners is a prerequisite for that shift to pay off, and opening submissions to any developer is one way to build it quickly. The comparison Google's own framing invites is to search engine optimization: just as SEO determined which websites surfaced in search rankings, participation in this program may determine which services are visible to users as AI agents take over more online tasks.

Google did not say when the platform would move beyond the limited preview. For app makers, the appeal is distribution to users who arrive with a specific task in mind, without building separate integrations for each Google product. For Google, each new partner deepens the utility of Gemini and Search at a moment when the company is trying to keep its assistant central to how people shop, plan and book.

The arrangement also leaves brand presentation and account authentication in developers' hands, a concession that may ease concerns about ceding the customer relationship. Separately on Wednesday, Google's DeepMind unit unveiled Genie 4, a new world model that generates interactive virtual environments from text prompts and is the first in the line to produce synchronized visuals and audio.

The company said game developers and filmmakers have begun testing it for prototypes and scene sketching, and that it is studying robotics training uses. The model remains experimental, with limitations in complex physical interaction, text legibility and keeping scenes consistent over long stretches.

Google said generated audio and video will carry SynthID watermarks. ‌ ‌ ‌ ‌ ‌ ‌