AI & Automation

Can your app live inside Google Search? Google opened its Connected Apps preview on 10/28 with 50+ apps already live: small businesses should first turn booking, lookup and ordering into an API, then think about Gemini and Search

2026.10.30 · 67 views
Can your app live inside Google Search? Google opened its Connected Apps preview on 10/28 with 50+ apps already live: small businesses should first turn booking, lookup and ordering into an API, then think about Gemini and Search
“

On October 28 Google launched a preview of its Connected Apps Platform, letting developers bring services into Google Search and Gemini through an MCP server, with more than 50 apps already live. We compare it with ChatGPT apps and Claude connectors, explain the opt-in limitation, and outline the API-first preparation Taiwan's small businesses can start today without an invite. ”

Share:

Short answer: Yes, but it is still a preview. Google opened its Connected Apps platform on October 28, and more than 50 apps can already be used directly inside Search and Gemini. Small businesses in Taiwan do not need to rush in; first turn booking, stock lookup and ordering into a clean API for about NT$80K–250K, then plug it into Google, ChatGPT or LINE.

On October 28, product leads for Google Search and the Gemini app announced the Connected Apps Platform preview on Google's blog. Third-party developers can now bring their services into Search and the Gemini app, so users can book a table, buy tickets or build a playlist without leaving the results page. Google says more than 50 Connected Apps are live across productivity, event ticketing and health and fitness; Adobe, Experian, Localiza and Peloton are live in Gemini and come to Search in the next few weeks. In one line: the results page is becoming a place to get things done.

Why now? Over the past year Google has pushed AI Overviews and AI Mode into more markets while looking for new ways to monetize. Alphabet's Q3 results showed search ad revenue of about $82B, up 10.5%, yet Google has acknowledged that ads in AI Mode currently earn less than traditional search ads. As users ask AI instead of clicking links, Google needs the "task completed" moment to happen in its own interface. Technically the platform uses the open Model Context Protocol (MCP), so one service interface can be understood by multiple AI platforms.

Google is not alone. OpenAI launched apps and the Apps SDK in ChatGPT last October, also built on MCP; Anthropic's Claude has a connector directory; Microsoft Copilot supports agents and plugins. The difference is the entry point: ChatGPT owns conversation, Google owns everyday search intent. The category is moving from "AI answers questions" to "AI operates services for you", and platforms compete on how many third-party services they can connect.

What does this mean for small businesses in Taiwan? You will not be in the first partner list, but the direction of "customers book, check prices and order inside AI" is set. Below: the details, actions for three readers, a platform comparison, the limits and a subscription-free way to prepare.

The details: an MCP server, the Connected Console, and Search, Gemini or both

What developers must do is concrete: expose an MCP-compliant service and submit it through Google's console.

  • Integration: a cloud-hosted MCP server, or Android MCP using the AppFunctions APIs.
  • Submission: register, test and submit in the new Google Connected Console, choosing Search, Gemini or both before review.
  • After launch: the console shows traffic, latency and error rates and manages brand identity such as verified logos and authenticated accounts.
  • Users: per support documentation cited by Search Engine Roundtable, searchers must add an app in their personalized settings to see it.
  • Availability: access opens in phases; regions, languages and pricing are not listed. Sign up on the Google Connected Apps developer page.

What three readers should do now: find the question customers ask most

Whether or not you get preview access, identify the single action most worth letting AI perform.

  • Brand owners and SMB owners: review a month of LINE and phone enquiries and count "is there a slot", "is it in stock" and "how much". The top one is your first API.
  • Marketers and SEOs: align your Business Profile, opening hours, price list and website. AI agents compare sources and rarely recommend a business whose data conflicts.
  • Developers and agencies: read the MCP specification once and estimate the hours to add an MCP layer to clients' booking or order systems.

Platform comparison: Google, ChatGPT, Claude and your own API

The technical base is converging; the difference is the entry point and whether users must opt in.

OptionEntry pointIntegrationFit for SMBs
Google Connected AppsGoogle Search, GeminiMCP server / Android AppFunctionsPreview, application required; users must add apps
ChatGPT appsChatGPT conversationsApps SDK (built on MCP)Open to developer submissions; suits conversational services
Claude connectorsClaude chats and workflowsMCP serverLeans towards work and internal tools
Microsoft CopilotWindows, Microsoft 365Agents and pluginsBest for B2B and internal use
Own API + MCP layerWebsite, LINE, any platform aboveLaravel API + MCP adapterAbout NT$80K–250K; no platform gatekeeping

What the announcement won't tell you: opt-in means less exposure than it sounds

It is pitched as a new front door, but there are at least three limits.

  • Opt-in friction: users must add apps in settings, the old "download our app first" problem again, so small brands are hard to discover.
  • Big brands first: the first wave is Adobe, OpenTable and Ticketmaster; there is no word on when local Taiwanese businesses or Chinese-language apps get access.
  • Platforms keep the relationship: if the transaction completes in Google's interface, you may lose customer data and remarketing, much like on delivery platforms.

A subscription-free alternative: API first, platforms later

The cheapest path is to make your core action your own API and treat platforms as just another outlet.

  • Weeks 1–2: pick one action (for example "check availability and book"), define inputs, outputs and errors, and write an OpenAPI document.
  • Weeks 3–6: expose the API from your existing Laravel system with API keys, rate limits and logging.
  • Weeks 7–8: wrap it with an official MCP SDK, test it with your own LINE bot, and submit to Google or ChatGPT when they open up.

At ScriptWalker (a Taiwan-based Laravel/Flutter custom development studio), adding this layer to an existing system typically costs NT$80K–250K over 6–8 weeks; new custom systems start at NT$80,000.

Frequently asked questions

Can Taiwanese businesses apply now?

You can request preview access, but regions and languages are not announced. Prepare your API first rather than rushing to submit.

What is MCP?

MCP is an open standard that lets AI call external services. One MCP interface works with Google, ChatGPT and Claude.

How much does an AI-callable booking API cost?

About NT$80K–250K and 6–8 weeks on an existing system, including docs, permissions, logging and an MCP adapter.

Will this affect my SEO?

Little in the short term. Users must add apps themselves, so rankings and your Business Profile remain the main exposure.

My take: it won't bring you orders for 18 months, but the groundwork is worth doing now

The mainstream reading is "Search is becoming an app store, grab a spot." I disagree: for 99% of Taiwanese SMBs, Connected Apps will bring close to zero orders in the next 18 months, because users must opt in, local markets are not open and the first slots go to big brands. What is worth doing is the requirement underneath: a stable, documented, machine-callable booking or ordering interface. That improves your LINE auto-replies and live stock checks today, and lets you plug into whichever of Google or ChatGPT opens first. For ScriptWalker, the service opportunity is an "AI-callable API package": add documentation, permissions and an MCP adapter to clients' existing systems so legacy software becomes infrastructure for the AI era on a small budget.

Sources

Share:
AI & Automation Back to Blog