Accessibility Adjustments

Use these optional tools to adjust reading and display preferences. These tools cannot resolve every accessibility barrier. Please contact the website owner if you need assistance.

  • Text adjustments
  • Content scaling 100%
  • Font size 100%
  • Line height 100%
  • Letter spacing 100%
  • Colour adjustments
  • Orientation adjustments

Google Connected Apps preview brings MCP services to Search and Gemini

Listen to this article

Google announced a Connected Apps Platform preview on October 28 for developers bringing services into Search and Gemini. Access is opening in phases, not as an unrestricted general release. Developers can choose which supported Google surfaces receive their integration before submitting it for review.

One platform, two integration routes

The developer guide recommends remote Model Context Protocol servers for cross-platform capabilities across web, iOS and Android. Android AppFunctions are the complementary route for actions that need local app state or deeper Android integration. These are different execution paths, not interchangeable packaging labels.

The registration instructions say each connected app has one connection type, which cannot later be changed. Its category also cannot be changed after creation. Choose these deliberately before investing in the rest of the setup.

What to prepare before applying

RequirementPractical implication
Public HTTPS endpointThe remote server uses Streamable HTTP.
One global endpoint per appSeparate data center URLs are not supported. Country-specific endpoints are limited to launches exclusively in that country or region.
Domain verificationHost the challenge response directly at the documented well-known path, without redirects.
Console permissionsAn owner or administrator accepts the program agreement. Team members need the appropriate Connected Apps permission.
Requirements from Google’s prerequisites guide, updated October 27.

The same guide requires public terms and privacy pages, an app icon, at least three useful capabilities and at least three example prompts. A working endpoint alone is not a complete submission.

A concrete test plan, not a tested integration

Consider an illustrative delivery service with separate tools for looking up a parcel and changing its destination. Test the lookup with a valid tracking number, an unknown number and an account that does not own the parcel. Then test the address change separately, including a user declining the confirmation. These are suggested evaluation cases, not results from a ByteForward deployment.

Google’s MCP guide calls for explicit tool annotations, including readOnlyHint for read operations and destructiveHint for destructive actions, plus typed arguments and clear descriptions. It recommends local MCP Inspector validation before testing the public endpoint in Gemini Developer Mode. Tool annotations are configuration signals, not proof that a workflow cannot make mistakes.

Preview access is not public distribution

The registration guide limits unpublished previews to added testers with Gmail accounts. Published tool definitions are versioned snapshots. Changing the server does not replace the published snapshot automatically. Updates require a new app version. Google reviews an integration before it can be published, and the developer publishes after approval.

MCP Apps UI resources have a separate beta caveat: they may not be available and can be ignored. Plan a useful tool response without assuming a custom interactive widget will render for every user.

What users can access now

Google says Adobe, Experian, Localiza and Peloton integrations are live in Gemini, with Search support planned in the coming weeks. Calm is also planned across both products. That partner rollout is separate from a developer gaining preview access or having an app approved.

The useful question is not simply whether an app supports MCP. It is whether its chosen execution path, account authorization, test coverage and release process fit the specific action a user wants to perform. ByteForward reviewed the documentation but did not register an app, deploy a server or test these integrations.

AI-generated editorial illustration by ByteForward. Google wordmark reference. Google is a trademark of Google LLC. This independent illustration is not endorsed by or affiliated with Google.

Maya Chen
Maya Chen

Maya Chen is focused on covering AI models, research, and the evidence behind new capabilities. Maya follows model launches, benchmarks, open weights, and scientific uses of AI with one question in mind. What changed, and how would we know? The voice is curious and exacting, with a soft spot for elegant technical ideas and little patience for a leaderboard without context.

Leave a Reply

Your email address will not be published. Required fields are marked *

Gravatar profile