September Demos and AMA: MCP, Conversational Generation, and Launch Lab's Winners

Every month, Bubble co-founder Emmanuel Straschnov goes live for a community AMA and product demos. September's session was about giving builders more control over how they build. He demoed an early look at MCP, plus a more conversational way to generate your first app and a handful of mobile updates covering testing and plugin development. From there, Emmanuel took community questions and closed out the session by announcing the Launch Lab’s winners.

Watch the replay below, or read on for a rundown of everything Emmanuel covered.

Everything Emmanuel demoed

Exact match search in the file manager (~1:00)

Emmanuel started with a quick update. The file manager in the data tab now supports exact match search. On apps with a large number of uploaded files, this makes it faster to find a specific file when you already know its name.

The mobile plugin editor moves to public beta, and BubbleGo is retiring (2:21)

The mobile plugin editor, previously in private beta, is now  available to everyone. It lets builders create plugins using React Native libraries, which removes most limits on what you can build for mobile on Bubble. 

As part of this same effort, BubbleGo is being retired from the App Store. Apple's current App Store policies don't allow apps that help people test other applications on their phone, and BubbleGo fell into that category. Two things replace it:

• A simulated device on the web preview runs your mobile app directly in the browser, using the same engine as a real device, so the behavior you see matches what end-users would see on their phone. This is available on the Free plan.
• Dev builds let you test on an actual device. They're built to Apple and Android’s requirements, so they take some time to generate, and they're available on paid plans with an app store connection.

A more conversational way to generate your first app (5:58)

Initial app generation used to be a single prompt that dropped you straight into the editor. Now it's more of a conversation. After picking an app type, the Bubble AI Agent (beta) asks follow-up questions to refine what you're building, for example whether customers should be able to message service providers directly, and shows a preview as it goes. 

You can also share style preferences, either by pasting your own color codes or uploading an image for inspiration, before generating the app. The idea is to front-load more of that back-and-forth so the first version is closer to what you need.

Editing directly from the live preview (8:55)

Bubble used to require two separate tabs to see your live app and edit it in the editor. Those are now merged. The preview button shows your app running in the same view as the editor, and you can ask the Agent to make changes, like changing a button's color, while looking at the running app rather than switching back and forth.

The Agent writes API calls, and now runs on Claude Sonnet 5 (11:23)

As a reminder for anyone who missed it a few weeks ago, the Agent can create API Connector calls on its own. Emmanuel demoed this by asking the Agent to set up an API call to Anthropic for a trail-finding app the team built for testing. He also confirmed the Agent now runs on Anthropic's newest model, Claude Sonnet 5.

Generated apps get animations, and mobile apps look more refined (13:08)

Emmanuel shared two related updates here, one on animations for generated apps and one on mobile apps looking more refined:

• Animations can now be included in AI-generated apps as part of the conversation with the Agent.
• Mobile apps get more visual polish from the initial generation, since the mobile team has improved how much styles get reused across a build. This matters in particular for app store submissions, since mobile app reviewers look at final polish in a way web reviewers typically don't.

An early look at MCP for editing your app (16:21)

Bubble's MCP (Model Context Protocol) integration is coming to beta within the next two to three weeks. Once connected, you'll be able to create and edit your app's UI and features directly from an AI tool like Claude or ChatGPT, without opening the Bubble editor, and do simple data exploration tasks like asking a model to summarize how an app is built or behaving.  

Emmanuel says, "Communities tended to be built around one tool, spending all their time in that tool because tools couldn't necessarily interact with each other. Now that MCP allows you to use intelligence from a different model to manipulate one tool, we see communities becoming more intertwined with each other."

Questions from the community

Workflow engine: branching and loops are close (~19:00)

Asked whether the upcoming workflow engine revamp would also improve backend workflows that expose apps as APIs or MCP servers, Emmanuel confirmed that branching (if/else) and loops (for/while) are coming, and that this will make backend workflows more capable for API and MCP use cases specifically. 

Is it possible to build a business using only Bubble? (~20:30)

Yes, largely, but Emmanuel noted you'll likely still connect to outside services through APIs for some things. For example, Bubble doesn't charge credit cards natively. That's a heavily regulated space, and Stripe already does it well. AI features are similar, as most builders connect to Claude or ChatGPT via API calls or plugins. He described this as core to Bubble's philosophy since the start. The platform doesn't restrict outbound API calls, so a Bubble app can talk to any external service.

AI usage limits, and using your own AI tool through MCP (21:48)

Some builders have started hitting new limits on AI usage in the editor. Emmanuel confirmed these are new, and something every product with an AI agent eventually has to introduce. Anyone who hits a limit should reach out to the support team, who can help case by case while a more permanent system gets built.

A related question was about using an external AI tool instead of Bubble's built-in Agent to avoid extra cost. While the Agent has been in beta, there's been no additional charge to use it. If you'd rather work with your own Claude or ChatGPT subscription, the upcoming MCP integration lets you connect it directly to the Bubble editor.

Emmanuel encouraged builders to keep using the Agent inside the editor, since it has more context on the app than an outside tool would. He also mentioned a second, separate MCP project, one that would let a Bubble app's own end-users interact with that app through their AI tool of choice. That's planned to follow the new workflow engine, with no date set yet.

Pre-rendering, for performance and SEO (~26:30)

Server-side pre-rendering is something the team wants, and people are actively exploring it. Emmanuel noted that Bubble apps tend to be more complex than typical no-code or vibe-coded apps, given how much logic they can support, so getting pre-rendering right takes extra care. No firm timeline, but a team is working on it, with performance and SEO both expected to benefit.

The Agent's models, data access, and mobile rollout (30:56)

A few quick Agent-related questions Emmanuel answered:

• Which AI models power the Agent? A mix. Claude is the largest share, alongside ChatGPT, open-source models, and some of Bubble's own fine-tuned models. The team picks models based on quality, cost, and speed for each specific task, since building UI, writing logic, and generating expressions don't all call for the same model.
• Does the Agent see your app's data, or just its structure? Both. Access to data and data types, including user data, helps the Agent debug and understand how an app behaves in production, not just how it's built.
• Is the Agent available for mobile pages yet? Not fully public. Bubble is close to opening it to everyone, including mobile, after a staged rollout meant to manage load and make sure things work well first.

Shorter answers

Lower-price dedicated plans aren't off the table. Emmanuel pointed to a base cost on AWS for dedicated infrastructure that isn't fully variable, which limits how low pricing can go. "If we see opportunities to do it, we will," he says (25:25).

Face ID or Touch ID for mobile apps don't have a timeline yet, but Emmanuel will flag it to the team as a recurring, reasonable request (33:00).

Support for Huawei's HarmonyOS isn't likely soon. Running a platform like Bubble's for China from the US comes with real technical separation between cloud infrastructure, and the near-term priority is doing Android and iOS well first (~33:30).

Building a social media app on Bubble is practical today, especially now that native mobile apps can be built and published to app stores (34:14).

Buying a plugin versus building your own with AI comes down to personal choice, Emmanuel says, tying it to a broader shift across plugin ecosystems as AI makes it easier for non-technical builders to build their own (35:23).

A one-on-one demo or walkthrough is available for Team-plan accounts and above, including Enterprise. For everyone else, Emmanuel recommends reaching out to a coach in Bubble's community, since coaches can offer this kind of support at a local level and in more languages than Bubble's own team could cover (37:02).

Offline functionality for mobile apps isn't a current focus. It matters most for specific use cases, like fieldwork without reliable service, and hasn't come up as a major blocker so far, though it's on the mobile team's longer-term list (38:05).

Theme-aware color variables for light and dark mode don't have native support yet. Bubble's own editor only added dark mode in June, and Emmanuel will raise the idea with the team (38:46).

A couple of builders also used the chat to thank the Bubble Support team directly. Emmanuel passed the thanks along, noting that positive feedback like this doesn't reach the team as often as complaints do, so it means a lot when it does.

Launch Lab cohort wraps up the AMA

Launch Lab was a free program that helps builders take an idea to a live app in a month, with weekly office hours from Bubble's team. Submissions were reviewed by a panel of judges, with apps spanning a wide range of use cases, from sports clubs to marketing campaign planning to compliance tools.

Emmanuel announced the winners (43:17):

• The Launch Lab Prize, the program's top prize, went to Jesper Vestergaard for KlubPlus, a platform for outdoor clubs in Denmark. Each club gets a fully branded website on its own domain, with event registration, member email, and admin tools all running from one Bubble app.
• Best web app went to Derek Duncan for Playbuk, an app that helps marketing teams plan and manage campaigns from first idea through execution.
• Best mobile app went to Emile Elie and Camille Köse for Business Class, an app that helps professionals build workplace skills through structured learning.

Emmanuel also recognized the other builders who took part in the program, thanking everyone for what they shipped. 

Full question list with timestamps

0:02 — Welcome and introduction to today's demos
~1:00 — Exact match search added to the file manager
2:21 — Mobile plugin editor moves to public beta, and BubbleGo is retiring
4:29 — Testing mobile apps with a simulated device and dev builds
5:58 — A more conversational way to generate your first app
8:55 — Editing directly from the live preview
11:23 — The Agent writes API calls, and now runs on Claude Sonnet 5
13:08 — Generated apps get animations
14:17 — Mobile apps look more refined
16:21 — An early look at MCP for editing your app
17:41 — Demo wrap-up, transition to community questions
~19:00 — Workflow engine, branching, loops, and apps exposed as APIs or MCP servers
~20:30 — Is it possible to build a business using only Bubble?
21:48 — People are hitting new AI usage limits
23:54 — Using an external AI tool instead of Bubble's built-in Agent
25:25 — Are lower-price dedicated plans something Bubble would consider?
~26:30 — Pre-rendering pages for performance and SEO
28:38 — Any updates for mobile responsiveness and native publishing?
29:35 — Excitement, and a firmer timeline, for workflow branching
~30:30 — Comment on recent AI industry concerns
30:56 — Which AI models power the Agent?
32:05 — Does the Agent see app data, or just structure?
32:29 — Is the Agent available for mobile pages yet?
33:00 — Any plans for Face ID or Touch ID on mobile apps?
~33:30 — Any plans to support Huawei's HarmonyOS?
34:14 — Is it practical to build a social media app on Bubble?
34:31 — Comment, thank you for the human support
35:23 — Should I buy a plugin, or build my own with AI?
37:02 — Is there support for a one-on-one demo or walkthrough?
38:05 — Any progress on offline functionality for mobile apps?
38:46 — Any plans for theme-aware color variables for light and dark mode?
39:23 — Comment, thank you for Bubble's human support and team
40:58 — Launch Lab recap and the cohort’s top prize winner
43:17 — Best web app and best mobile app winners, and closing remarks