Vibe coding with Gemini: the tools, and how to ship the result

TL;DR. Vibe coding with Gemini happens on four surfaces as of August 2026: Google AI Studio's Build mode (full-stack apps in the browser), the Gemini app (Canvas for small apps, Spark for agent tasks), Antigravity (Google's agent-first IDE, desktop app and CLI) and the open-source Gemini CLI (now limited to API-key and enterprise use). To ship the result you have three paths: AI Studio's own Cloud Run deploy, handing the project to a Gemini surface connected to Hatchable over MCP so it deploys to a live URL with a database included, or importing a static export at /deploy.

What "vibe coding with Gemini" means in August 2026

Vibe coding is describing the app you want and letting the model write it, and Gemini vibe coding is that loop on Google's models. The confusing part is that Google now ships it on several surfaces with different strengths, and most of them stop short of the step people care about most: the app running at a URL other people can open. This guide covers what each surface does, then how to get the result live. If you want the general idea first, read what vibe coding is.

One dated line applies to everything below: descriptions of Google's products are based on Google's public docs and announcements as of August 2026, and these products change monthly. Check the linked pages for current details.

The Gemini app builder surfaces, briefly

Google AI Studio Build mode

Build mode at aistudio.google.com is the closest thing Google has to a dedicated Gemini app builder. You type a prompt and, per Google's docs, "AI Studio creates a complete application": a React frontend by default plus a Node.js server side for API calls, database connections and npm packages. As of 2026 the coding is done by the Antigravity agent, Firebase Firestore and Authentication can be provisioned automatically, and your Gemini API key is injected server-side so it never ships to the browser. There is also an Android target that generates a Kotlin and Jetpack Compose project.

What it is genuinely good at: prototypes that call Gemini, because the key handling and the model wiring are done for you. The trade-off: the output is a Google-shaped stack (React, Node, Firebase, Cloud Run), which is great if that is where you want to live and more work to move if it is not.

The Gemini app: Canvas and Spark

The consumer Gemini app has two relevant pieces. Canvas builds small apps and pages from a prompt; Google says Canvas apps can keep data between sessions and be shared by link. Spark, announced at I/O in May 2026, is Gemini's always-on agent: you describe a task and it runs it, including through Connected Apps. Custom MCP servers can be added as connected apps too, though as of August 2026 Google's help pages limit that to adults in the US on personal Google accounts. Spark is what lets the Gemini app hand work to an outside service such as a host.

Antigravity

Antigravity is Google's agent-first development environment: an IDE (a VS Code fork) at launch in November 2025 and, with version 2.0 at I/O 2026, a desktop app that runs several agents in parallel, an Antigravity CLI and an SDK. It is the right surface once the vibe-coded app has become a real codebase you want to keep editing locally. As of August 2026 Google offers it free to individuals in public preview with usage quotas that have been adjusted more than once, and higher quotas come with Google AI subscriptions; treat that as provisional. It supports MCP servers, so it can talk to a host directly.

Gemini CLI

Gemini CLI is Google's open-source terminal agent (Apache 2.0, on GitHub) and it supports remote MCP servers with OAuth. One thing to know before you pick it: Google announced in May 2026 that Gemini CLI would stop serving requests for free, Google AI Pro and Ultra accounts on June 18, 2026, and pointed those users at the Antigravity CLI. Per the same announcement it remains usable with paid Gemini API keys and under enterprise Code Assist licences, and the repository stays up unchanged. If you have a paid API key it still works, and the MCP setup is one line:

gemini mcp add --transport http hatchable https://hatchable.com/mcp

How to ship it, path 1: Google AI Studio's own deploy

AI Studio deploys to Cloud Run. Per Google's deployment docs, there are two tiers. The Google Cloud Starter tier "lets you publish up to 2 full-stack applications without setting up a Google Cloud project or billing account", in a single Cloud Run region, with a subdomain under ai.studio. Google says people with active or prior Google Cloud billing accounts, and Workspace enterprise accounts, are not eligible. Beyond two apps, or for anything more, standard deployment needs a Google Cloud project with billing enabled. You can also download the code as a ZIP or connect a GitHub repository for two-way sync, and the Build mode docs note that when you share an app, its Gemini API calls count against your usage and paid models can cost money.

This is the shortest path if a Google Cloud account is fine for you. The full walkthrough, including what to do when the starter tier says no, is at how to publish a Google AI Studio app.

How to ship it, path 2: hand the project to an AI connected to Hatchable

Hatchable is the platform where AI-built apps go live. Instead of a dashboard, it gives your AI an MCP server at https://hatchable.com/mcp (OAuth sign-in, no API key). Once connected, the AI creates a project, writes the files, deploys, and hands back a live URL at your-app.hatchable.site. Every project includes its own private Postgres database, file storage, sign-in for your app's users, email, scheduled jobs and AI calls on your own provider key, so "add a database and a log-in" is a sentence, not a provisioning session. Three Gemini surfaces can connect:

1. Gemini (Spark)

   Add https://hatchable.com/mcp as a custom app under Spark, Connected Apps. Six screens, two of them security notices. Then describe a task, for example "build me a countdown site on hatchable". Every screen is shown at connect Gemini.

2. Gemini CLI

   Run the gemini mcp add line above, approve the browser sign-in, then ask it to build and deploy. Details and troubleshooting: how to add an MCP server to Gemini CLI.

3. Antigravity

   In the Antigravity chat, type "Add the https://hatchable.com/mcp MCP connector with OAuth." The agent writes the config, you approve the tab, and it can deploy from the same session. See connect Antigravity.

If the app started in AI Studio, open the ZIP or the synced GitHub repository in Antigravity (or point Gemini CLI at the folder) and ask it to port the app onto Hatchable. The usual job is swapping the Firebase pieces for Hatchable's database and sign-in, which are SQL migrations and a config line, and re-pointing the Gemini calls at Hatchable's AI call with your own key. Hatchable is not a long-running-process host: server code runs as API route files, so the agent rewrites the Node server's routes into that shape rather than running it as is.

How to ship it, path 3: import a static export

If what you built is front-end only (a Canvas page, a static build out of AI Studio), the /deploy page imports a ZIP, a project folder or a GitHub repository. Static sites import directly and go live; anything with a server side is handed to your connected AI to port. The free plan publishes one app to the open web with no card, and Builder is a flat $12 a month for unlimited published apps.

Which path to pick

General characterisations based on public information as of August 2026; check each vendor for current details.