Google has opened a front door for developers who want their apps to appear inside its AI products. On October 28, 2026, Google announced a preview of the Connected Apps Platform, "a unified way for more app developers to bring their capabilities directly into Gemini and Search", written by Chips Mistry (Group Product Manager, Search) and Mai Lowe (Group Product Manager, Gemini app) in Google's announcement. Build once, register in a console, and choose whether the app lives in Search, the Gemini app, or both.
Connected apps are not new. Google introduced them earlier this year and we have covered pieces of them, such as the Gemini small business hub with Shopify, Canva and Xero connections and Gemini Personal Intelligence on Android Auto. What changed today is the supply side: any developer can now apply to build one, with a console, metrics and a single integration path.
TL;DR: what Google announced
| Question | Answer |
|---|---|
| What is it? | A platform to build and manage apps for Search and the Gemini app |
| How do you integrate? | Cloud-hosted MCP server, or Android MCP via the AppFunctions APIs |
| Where do you manage it? | Google Connected Console: register, test, submit, then track traffic, latency and error rates |
| Where can the app appear? | Search, Gemini, or both, chosen before review |
| How many apps exist? | More than 50 live across Search and Gemini, per Google |
| Who can join? | Preview; TechCrunch says a US developer waitlist; Google says more markets later |
| Docs | developers.google.com/connected-apps |
What is a connected app, in plain terms?
A connected app lets you do a specific thing with a service without leaving the AI conversation. TechCrunch gives examples already shipped: designing a flyer in Canva or building a playlist in YouTube Music. Google's post cites ordering groceries and creating playlists, and shows a diagram of Search handing off to Gemini to confirm a dinner reservation.
The idea is an assistant that calls your app's tools on the user's behalf, instead of the user opening your app. If you have followed the connector wave, this is Google's version of what OpenAI and Anthropic ship; see our guides to ChatGPT Sites hosting MCP servers and plugins and Claude connectors for Gmail and Drive, and the top Claude connectors for comparison. TechCrunch also notes that ChatGPT and Meta's Muse have tried similar ideas.
How do developers connect?
Google offers two paths, both built on the Model Context Protocol (MCP), the open standard for exposing tools to AI clients. If MCP is new to you, start with our overview of open and closed MCP servers.
- Cloud-hosted MCP server. You expose tools and authorized data from a remote server. Google's docs link a remote MCP server guide and an OAuth 2.1 guide for authorizing the Gemini app to act for a user. The docs also mention MCP Apps for interactive interfaces; for a sense of what that looks like, see Microsoft Flint's clickable-chart agents built on MCP apps.
- Android MCP through AppFunctions. This is the client-side route: an Android app shares functions with assistants such as Gemini, and tool execution happens locally on the device. It is built with the Jetpack AppFunctions SDK. Our write-up of Android Studio Rabbit 2 and its agent APIs covers the surrounding Android tooling.
Then you work in the Google Connected Console, which during the preview is the self-service onboarding route. It handles tool registration, usage analysis, and the review and publishing process. After launch you can track traffic, latency and error rates, and Google says developers keep control of their brand identity, including verified logos and authenticated accounts.
Who is live, and what is coming
Google says it has launched more than 50 Connected Apps across Search and Gemini, spanning productivity, event ticketing, health and fitness. Named integrations:
- Live in Gemini now: Adobe, Experian, Localiza and Peloton. Google says they come to Search in the coming weeks.
- New partner: Calm will launch across both Search and Gemini.
- Already in Search, per TechCrunch: Instacart, Canva, YouTube Music and Yahoo Fantasy Football, alongside Google's own services such as Photos, YouTube Music and Google Home inside Gemini.
Rollout is phased. Google says it will add more developers, markets and Google products over the coming months, and TechCrunch reports Google did not give a date for broad availability.
Why this matters for builders
The pitch is distribution. A single integration reaches two surfaces: the Gemini app and Search, where AI answers increasingly end in an action instead of a list of links. TechCrunch frames the opportunity as comparable to how SEO shaped website visibility, and argues it only works if enough apps are connected. We read that as the real bet: the more tool calls an assistant can make, the more it becomes the place users start tasks.
For a small developer, that cuts two ways. Upside: your app can be invoked at the moment of intent, with Google's reach. Downside: the user relationship moves one step away from you, and ranking among connected apps will be decided by rules Google has not yet published.
Compare the retail side of this shift. Walmart is already putting an agent gateway with MCP and spending limits between shoppers' agents and its catalog, while Amazon blocks agents. Google's route is the platform model: be the place where businesses opt in.
What is not confirmed
Be careful about what the sources do not say. In the pages we read:
- Review criteria and timelines are not described; Google's docs mention a review and publishing step but give no steps.
- Policy terms sit on a separate Policy page we did not see reproduced in the announcement.
- Eligibility is not stated in the documentation; the US waitlist detail comes from TechCrunch, not from Google's post.
- Revenue and payments: nothing about fees, commissions or how purchases made through Gemini are handled.
- Search availability dates: "coming weeks" for the Gemini apps, with no specific day.
Treat anything beyond that as speculation until Google publishes the policy and pricing pages.
Security and permission questions to ask before you connect
Any system where an AI assistant acts on a user's accounts raises the same questions we have raised for other agents. What scopes does the OAuth grant request? Can the assistant take an action, such as a payment or booking, without a confirmation step? What gets logged? Our write-ups of MCP rug-pull and tool-poisoning attacks and the Grok Bot vault that lets an agent use saved cards show how this goes wrong when tool descriptions or permissions are too loose.
If you run agents that touch real systems, AgentBeam is the agent security platform from the explainx.ai team that stops AI agents before they take dangerous actions. It is one option for adding a check between an agent and a risky action.
A practical checklist for developers this week
- Decide if you want the surface. Ask which two or three user tasks your product could complete in a chat or search answer, and whether you would be comfortable with users never opening your app for them.
- List your tools. An MCP server is a set of named actions with inputs and outputs. Write them down as plain verbs: search listings, create a booking, fetch an order status. Keep write actions few and clearly separated from read actions.
- Plan authentication. Google points to OAuth 2.1 for letting Gemini act for a user. Use narrow scopes and short-lived tokens.
- Apply for the preview at the documentation page and expect a waitlist.
- Decide your target surface. Search, Gemini or both is a choice you make before review, so know which audience matters.
- Prepare monitoring. The console reports traffic, latency and errors; set your own alerts too.
How it fits Google's wider pattern
This is the third piece of a pattern we have tracked through October: Gemini Skills replacing Gems, personal context features on Android, and now an app ecosystem. Together they point to Gemini as an agent layer that can read your context, run reusable skills and call third-party tools. The open standard underneath, MCP, is also being stewarded across companies; see our coverage of the Agentic AI Foundation's ten projects.
The competitive question is whether developers will build once for MCP and reach every assistant, or maintain a separate listing per platform. Google's use of MCP as the integration path is a point in favor of the first. The console, review step and brand rules are the part that is Google-specific, and that is where lock-in would come from.
What users should know
If you use Gemini, expect more apps to show up as suggestions and actions. You will normally be asked to link an account before an app acts for you. Review what you grant, link only the services you use, and revoke access in your Google account settings if you stop using one. Nothing in the announcement changes your existing settings, and the details of how the review protects users have not been published.
What happens next
Google says Gemini's connected apps for Adobe, Experian, Localiza and Peloton will reach Search in the coming weeks, that Calm will launch on both, and that more developers and markets will be added over months. We will add an update section here when the policy page, eligibility rules or a non-US rollout is published.
Related reading
- Gemini small business hub with connected apps
- Gemini Skills replace Gems
- ChatGPT Sites host MCP servers and plugins
- Top MCP servers, open and closed
- Walmart agent gateway with MCP and spending limits
- MCP rug pulls and tool poisoning explained
- Official: Google's announcement and Connected Apps documentation
Details are accurate as of October 28, 2026 and come from Google's announcement, its documentation and TechCrunch's report; availability and terms may change.