Cursor for Mobile Developers

0:00 Cool. Let's dive in here. Okay, I should share my screen. I'll assume you can see my screen. Let me know if you can't. Um yes, so today uh we're going to be diving into uh you know, how you can get an effective use of cursor for your mobile environments, right? So, uh rough agenda here is we're going to be looking at uh first off uh an agent window, which is uh a new uh desktop application we released a few weeks ago. And the approach we're going to take

0:32 today for mobile development is using that agent window as a sidecar to help enhance both iOS and Android um IDEs and kind of have the integration flow uh between the two in a bidirectional way. So, I'm going to switch over to a cursor here. Um so, to kick to kick off um I'm going to walk through something we released called the agents window. Um and um you know, many of you have you know, have probably experienced cursor through uh the IDE, right? Which is uh VS Code fork um that has evolved a lot over the past

1:08 few years. And we uh we took an approach in building kind of a ground-up uh native application that is kind of first principled around the agentic experience and and coding with agents. So, when you're in VS Code here, uh you'll be able to do either uh command shift P and you can do uh view and switch to the agents window. There's also going to be a button on the top right where you can click agents window and that'll pop open agents window.

1:41 So, like I mentioned, agents window, you know, is an agentic first experience and we're going to be using this to uh you know, bridge the gap between cursor and and different IDEs. So, I'm just going to walk through kind of the layout of of agents window before we jump to the mobile side of stuff just uh kind of level set on uh the different features on here itself. Um so, you're going to have your your main your main uh prompt window. I'll be able to kick off agents

2:06 for me here. And if you click on the top left, this is going to be your agent pane. So, you'll be the ability to open up multiple workspaces, multiple uh repositories uh within this one instance. And you can run agents across those different repositories itself. Um each uh kind of workspace you open will contain its own uh indexing, right? So, it's going to be kind of co-located to that workspace itself. So, this kind of allows you to run, you know, multiple agents across multiple projects, um

2:36 switch between the progress of them, and kind of see them unfold uh in real time. And something cool about the agents window, if you are running cloud agents, you'd be able to uh also see and kick off cloud agents from this window itself. Um won't be going into that in in this session, um but um I think there's some other ones on the workspace uh workshop page, which which you could kind of dive into.

3:00 And then on the right of the agents window, you're going to have uh the right side panel. And the right side panel is going to be, you know, the area where you can get the more uh traditional uh inspection investigation, yeah, into the code repository itself, right? So, you have your uh traditional file browser here. So, yeah, you're able to kind of go in, uh see files I have uh been bindings um kind of connecting them there. Um so, you can make any edits you

3:27 need there. And then you're just going to have your uh terminal here. So, for each agent that's going to run, it's going to have its own sandbox terminal. Um so, you can access that as well. And then you're going to have that cursor browser in here. And the cursor browser uh is a way to kind of extend browser functionality to the cursor agent. Uh we find this to be incredibly powerful where uh the cursor agent is able to tool call, and it can test its

3:52 changes it does um within this integrated cursor browser itself. And the last piece of this right side panel is going to be this kind of review pane. Right? So, as agents work, they're able to review the work that that it does. And you can actually create branches and commit on that branch itself from this pane itself. Cool. And the last piece of agent window just want to touch on real quick is we have this marketplace area where you can come in here and install different plugins. And plugins will provide we can

4:25 open one up here. Um say this Amazon Amazon location service is going to provide, you know, a package of MCP servers and skills that if you click add to cursor, it's going to bring it down install those MCP servers skill configuration files and your cursor instance or cursor agent will have access to to those skills and MCP servers and be able to use that while while while you're interacting with it.

4:53 So, I bring this up because we're going to go over an IntelliJ plugin which helps with Android Studio development, but we'll come back to to that one momentarily. Cool. That that was a quick run through of the agent window. And we'll switch over gears now to the mobile side of stuff.

5:12 So, we're going to start off with iOS development. And you know, with with mobile development with iOS and Android development, right? You know, you can of course generate Swift code or, you know, Kotlin code. You know, with cursor, but we find the big missing piece of the integration is the ability to, you know, run the simulator, run any debugging, get build logs from the IDE itself. So, we found a few ways to be able to expose that through these integrations. And on the Xcode side,

5:47 we're using an MCP integration that Apple actually maintains itself. So, what Apple did is within Xcode, they spin up an MCP server that exposes a list of different tools in these categories. Um and through that in our agent window, we'll be able to interact with these tools and kind of you know, both develop and generate code, but also use these tools to build test and get you know, finer diagnostic information of what's going on. Um some interesting tools

6:21 within that which I find can be quite powerful is it does have a really good documentation search search to Apple's SDK documentation which can be really powerful for the kind of the finer pieces of mobile development. So, that is exposed in the tool itself. And one of my favorite ones of what's exposed is this render preview tool. And this will actually give screenshots of the simulator of the app running and give that to the agent. All right, so

6:53 the agent would be able to make changes, rerun the build, and then get actual screenshots of what the UI looks like and kind of use that to continue to iterate to get the expected end goal that that you would you would want. And the last cool one is it does also expose a rebel environment that the agent can use to interact there.

7:17 So, let's go into the setup of this. There's the first step is going to be setting up the MCP configuration in cursor. So, you have a few options here. I will actually drop this in the chat. So, we can all use that. And what this will do is it's going to add this configuration into your user directory mcp.json file. All right, so [clears throat] for those of you unfamiliar with mcp or mcp.json file, mcp stands for model context protocol and it's a way to expose, you know,

7:59 tools and functions and APIs to LLMs and have have the ability for LLMs to invoke and use those tools during its agent run, during its agent trace. So, Cursor has an mcp file definition at your user level. So, if you're in your home directory, you can go to dot cursor. Uh dot cursor.

8:23 It's going to be mcp.json. And in here is where you can list the mcp servers. So, with this configuration, you can you can either add the JSON itself or you can use this deep link file, which I'm going to use that one. Which will automatically set it up deep linking into the agent window and then this will kind of bubble out into the mcp.json itself. So, we can go ahead and click install there.

8:50 Cool. And you noticed once I clicked installed, you have Xcode kind of bumping up there. And so, that is already starting the communication over the bridge with Xcode, right? So, on Xcode side, you're going to come up you're going to have to come in here and click allow, which will open up that bridge between Cursor and Xcode for that mcp server to work. It's kind of annoying, you're going to have to do this every time you open up Xcode, but it's I think it's like five allows and then and then you should be good there.

9:26 So, once mcp is set up, kind of how this works Yeah, it was a lot of allows, but that is Apple for you. The way it kind of works under the hood is we have Xcode which uses its XPC protocol and exposes the tools through there through this MCP bridge and then on the cursor side we have the MCP client which is going to you know use that MCP server to get to get what tools are available and make the two calls itself. And during your guys setup process if you find that you run into this this error which we saw

10:03 some some some team members run into you can run through uh these and maybe I'll just post this in the chat as well. Uh these commands here which should uh fix that issue um but other than that it's usually pretty straightforward to to kind of get that running.

10:26 And we can take a look now what this will look like. So I have Xcode running here and I'm just move my windows around a little bit. So you have Xcode running and then we have our simulator. And we're running a little demo app. It's a Snapchat clone right so you kind of have all your options here. Um it's running the the build and and the simulator of course through Xcode itself right and we're going to sidecar the agent window right so when I mean sidecar it really means just running it

11:06 right next to it right so kind of get be able to jump back and forth between both IDEs. And let's go through a small change in the simulator. So we have these lenses here right and let's ask cursor to uh Oh yeah thanks Alex. So let's ask cursor to um you know instead of having all these buttons laid out you know on this home screen itself. Let's have a single lenses button that when it's clicked, it's going to open up this whole tray of lenses you could kind of cycle through.

11:39 But, we're going to do that through the agent window here. So, I'm going to go in here and I'm going to say and I'll use composer to fast so you you can kind of see this happen quicker. Can you update the lenses tray to be a button. Oops. All right, it's a usual kind of cursor interaction. Um it maybe be beneficial to have this to be a plan, but just for the sake of time, we'll we'll uh kick off an agent straight up here.

12:18 And you can see it's it's exploring through uh the files here. It's making the change. And and then it's going into into the building and compiling of uh the uh the binary itself, right? The Xcode build itself. And I actually have a skill in here which will do a refresh and run. So, I'm going to go ahead and invoke that so you can kind of see what exactly how that would be.

12:45 Let me just fire that off. So, I have a specific skill um that I built that will uh instruct it to use the MCP tool to to go ahead and and run this and rebuild it. So, we can see that going on there. I'll see that come up and it actually looks like it's using the terminal commands. So, that probably means my MCP server is not fully hooked up. So, maybe I'll go do a quick refresh a reload of this itself so you can see that full functionality. Um so, again, I'm doing command shift P

13:22 to get that kind of quick action menu and it does look like it's coming up. Um so, we'll let that come up. Uh, we'll do a quick reload here. It goes. It did come up. And then let's see another refresh to kind of see that MCP tool action there. But we do see our change has has persisted. So we have uh, our lenses button in there um, that when clicked pops up the tray of of lenses there. So, this is kind of like one way, you know, we see uh, you know, kind of having a deeper integration

13:55 between the two um, and kind of extend the functionality. Um, if you guys are are CLI users, you can you can also use uh, you know, the CLI here as well. Um, for those of you that aren't familiar with the CLI, uh, we have a CLI that matches the same uh, functionality of uh, of of the IDE and of this agents window. So you're able to do, you know, debug mode, plan mode, and kind of the whole the whole gamut of stuff as well. Um, but we find that, you know, interface-wise the agent window can be

14:23 really uh, really effective on kind of this sidecar mechanism. And uh, kind of like, you know, Cursor roadmap-wise, um, for those of you that are on cloud agents, um, we are exploring, you know, how we can run, you know, simulators in in cloud agents for both Android and iOS. So uh, there's more on that to come in the future to kind of, you know, make this integration, you know, even more kind of um, in-depth and a little bit more fluid than using this MCP tool to uh, to talk across the the two apps there.

15:00 Cool. And we can see here the work it did uh, where it built a debug path, uh, installed and launched the app, um, and then we have that that change there. And do we have more questions about allowing the MCP tool? Cuz I like I mentioned, I restarted Cursor, so Xcode's going to ask for that.

15:21 Um, so I'll pause there for a second If there are any questions specific to this iOS integration. I guess Alec, I don't know if there were any questions or I can continue to the Android piece. >> Um there was actually one on Agent's window um when you were initially showing it just around like how you can see what branch you're working on. Is like an easy way to switch between branches? >> Yeah, that's a great question. Let me open this up here. Um so I got new thing. So, you'll be

15:49 able to see it uh in a couple places. Um so here um you're going to see what branch you're on uh on the top right here. And then if I start a chat uh or actually if if you if you have changes, so maybe you can see some changes here. You'll see it on the review tab. Let me open that up. Um on the top left there as well.

16:18 So, you can uh also see the kind of commit and push button on the right here. So, if you were working on say I'm working on the main branch and I made changes but I actually want to push that to the main branch itself, then if you click that arrow button on the right next to commit and push, you can uh create a new branch and then and then commit on on there as well.

16:37 >> And then Sunny Pranav asked like, "What if I wanted to build an iOS and Android app in parallel? Like what could that look like?" >> Yeah, for sure. Um you you totally can and you can do it through one Agent window itself. Um so, we can go through the Android part but um yeah, it's kind of very similar of adding these MCP tools uh into into Cursor and then you'd be able to uh kind of run multiple agents across multiple projects, right?

17:02 So, um we're going to have this iOS project open here. We're going to open up the Android one here. And then you can just run agents in in either one um to to kind of kind of build uh build them both in parallel. >> And then, yes, see uh Martin, is it possible for an agent test for an agent test of speech in the simulator? Like if they could figure out where to click on the simulator to anticipate future flow. Can an agent interact with the simulator? Yes, that's a good question. So, as of

17:35 right now, uh it can't, right? So, it can it can bring it up, um but it can't click around within the simulator itself. Um it can take screenshots. Um so, if you already have kind of the screen up and it's uh refreshing the screen itself, they could take a screenshot of that with that render preview MCT tool, but it can't it can't click around uh quite yet.

18:00 And then, bro, how can we test it without Xcode? Uh yeah, it's a good question. So, if you uh want to write like more traditional, you know, unit test itself, um you can obviously do that through through cursor. Um the simulator itself is run through um as of now, the best integration is through iOS's kind of uh Xcode's infrastructure. Um so, that's that's probably the best way to to handle that itself.

18:30 And cool. And then, yeah, I'll answer these a couple last ones. Uh All right, what are the two benefits of MCT and having Chris on the code in the terminal? Um yeah, so, the the big biggest benefit is for that building and running. Um and when you build uh the app, you're going to get the uh one runtime logs, which is a little bit more or build time logs, which is a little bit more hydrated from Xcode versus the build command in the terminal. Um so, that's probably the the

19:03 biggest one I would I would say. And Pranav, we'll use this as the last one. Last question before we switch to Android. Is there any alternative to MCT if it's expensive? Uh yeah, I mean there's ways around it. Like you can uh for Xcode itself, the the way they expose it is is through MCP.

19:21 You could wrap it in like a CLI tool locally and then call the CLI itself, but what I'd actually recommend on the expense side is, you know, whenever you're using uh heavy MCP tool calling for, you know, a task or a prompt, uh don't don't use a high reasoning model, right? Don't don't use a heavy Opus high thinking uh use composer two, composer two fast, you know, more of an execution model, which can keep the spend more under control um versus kind of burning tokens calling functions and then waiting for a

19:51 feedback there. Cool. So, let's switch gears to uh Android Studio here. So, I'm just uh set there. Cool. So, switching to Android Studio here. Slide this. Um in the Android Studio, we're going to have uh two different options for for integrations. Uh the first one is going to be using a plugin called JetBrains AI Assistant, um which will let you use cursor within Android Studio uh through something called Agent Client Protocol. So, Agent Client Protocol is an open standard for coding agents, and it lets, you know,

20:43 different IDEs or, you know, really any client implement that spec. And if it implements that spec, then any coding agent that implements uh the spec as well will be able to be integrated kind of off the bat into uh that IDE or software itself. So, um JetBrains has uh support for ACP. Um I I like uh Neovim and Emacs have some uh open source work that's being done for for ACP integrations. And there's also ACP integrations into, you know, non-coding related IDEs. So, for those of you that use Obsidian for notes, you

21:22 can actually have, you know, Cursor be in your Obsidian you know, ID IDE in your Obsidian editor to, you know, help you and have an agent as you write your notes. >> I'm going to switch over to Android Studio here. And the first thing you want to do for this integration is you want to get the plugin. All right, so you're going to come in here. You want to make sure it's this JetBrains AI Assistant one.

21:52 So, you want to go ahead and install that. And the second part for this, if you don't have the CLI installed, you're also going to want to install the CLI or actually maybe Alec if you want to throw that throw that link in. So, you want to install the CLI cuz the CLI agent will run that ACP bridge. So, you're going to need that binary installed locally. But, once you have those two installed, let's see. Make sure we don't need a reset or restart.

22:24 So, we have that installed. Then if you come back to Android Studio, you're going to see this AI chat option open up. And if you click into that, you're going to have a bunch of options for agents. And if you go to install from ACP registry and look for Cursor, you have the ability to install that Cursor agent. And once you install the Cursor agent, it's going to either prompt you to authenticate within this side AI chat or you can authenticate on the CLI and that will propagate up into

22:58 this Cursor instance itself. And once that is kind of completed and done, then you're going to have uh the full kind of, you know, cursor agentic experience but now, you know, fully integrated into uh Android Studio, right? So, what that is going to look like is um you know, you have you have the ability to, you know, pull in different things into the context window. Um and then also use Cursor's different agent modes to kind of, you know, get the functionality you want. Um I actually don't see it popping up here,

23:28 so I'm just going to quickly restart uh Android Studio to see if we can get that showing. And uh you know, with with ACP integration, right? Um as as the spec kind of evolves, you know, uh Cursor will kind of stay hand in hand with that spec itself, right? So, um it looks like for my local right now, the uh the agent modes aren't aren't popping up. So, let me let me know if you guys are experiencing the same thing as you go through this, but um unlike JetBrains' IDEs, um which is uh similar

24:09 to Android Studio, um you'll have access to not all the agent modes, but you'll have access to ask um like regular agent and then plan. Um so, uh you won't have access to the debug mode, so that would have to be uh if you would have to use that in the CLI or the agent window itself.

24:25 But everything else will work kind of similar here, so I'm running I was running uh a little Google Keep note-taking app, right? So, you come in here and um kind of request information from the agent, so I'll ask it to tell me about this app.

24:42 And uh you're going to see Oh, yeah. Oh, there we go. It's asking us to log in, which makes sense. That's probably why it wasn't popping up, so um I'm going to click that login button. Uh we can OAuth and then return to Android Studio. And then Okay, that makes sense. Now now we're seeing our agent modes. So, we'll see that agent plan and ask, like I mentioned, debug isn't going to be exposed here, but once ACP spec supports debug, then we'll be able to to

25:07 propagate that out. And then, you'll also have the ability to, you know, select which model you want to use. So, whatever model definition you have set in the CLI, then that's going to also get propagated out to here. And everything else will be very similar. You're going to see the agent trace. So, as it kind of thinks through this question here, and maybe I'll start a new chat cuz I just logged in.

25:36 So, as it kind of thinks through this, we're going to see its thinking process. You're going to see any tool any tool calls, any terminal tool any terminal commands being run. It's going to be a similar allow list process as it as it comes through that. So, it should shouldn't mirror, you know, your your regular kind of experience that you would have in the CLI or elsewhere. I'm not sure what's going on there though. Let's see.

26:03 Let's try 5.4. I don't know if you guys have heard, but there's a lot of rumors that GPT 5.5 is going to be released in in a couple hours. So, that will be exciting to see. Cool. And let's see what's going on here. So, let's just take the settings real quick.

26:31 Something else to keep in mind too is, you know, you have the ability to uh specify MCP servers at the Android Studio level. And if you are doing that and you want to expose that to cursor as well as a tool call ability, you can enable this pass to IntelliJ MCP server, uh will which will pass that along. Go.

27:08 And the kind of second option here for Android Studio is something similar to how Xcode is set up how the Xcode kind of integration set up with those MCP servers. So, if I go back to Android Studio, um and we're going to go back to the plugins area. Until I see Xcode asking me to allow things again.

27:31 So, uh we go back to plugins area, um there's a a suite of plugins that you can actually enable, which will uh expose certain functionality from the IDE itself. Um so, the ones I would recommend is there's an MCP server one, which is going to expose um a bunch of useful functionality. Uh In addition to that, there's a debugger MCP server, which is going to uh install uh ins- once it installs, it's going to expose uh again over an MCP server um even finer control of

28:12 debugging mechanisms. Yeah, so the ability to set line break points, um conditional break points, uh start uh step out, step in, kind of like, you know, anything you would probably do in the UI, you'd be able to do through this MCP server itself. So, that's that's also a fun one.

28:30 And uh Okay, cool. It looks like our our chat got there got kicked off. And then, also the kind of a similar setup and we can kind of just see how this runs is uh on the cursor side, you know, you can kind of do that side car. Bring that over. And go through the installation process itself. I actually have and I was talking about plugins earlier cuz I actually have an IntelliJ plugin that I uh I use for, you know, IntelliJ based development. And this IntelliJ plugin

29:09 wires in the MCP client definitions for the servers we just added automatically. And then it also has a bunch of skills that I added that kind of helps steer and and guide cursor to, you know, use those MCP tools effectively and talking efficiently. So, I already I already have this installed. I can definitely share this with you guys if that is something that is interesting.

29:36 But we'll just go through a reinstall here so you can see how that works, right? So, again, add their cursor. You can choose to add it to your user directory definition or just the project itself. So, I'm I'm going to just add it for for myself there.

29:51 And once we have that, you know, we'll we'll be able to interact in a in a kind of similar manner. So, if we just open up my project here. And I'm actually forgetting where I put that. Just give me one second here. All right. So, then of course just like kind of the other side, you know, you can either use this ACP or you can use this and it with skills and and rules, you know, it'll kind of wire in pretty well to make changes into this Android app itself. And uh it's like Xcode again is asking

30:53 for more permissions. So, I can go in here and say, "Can you Let's build." I'm actually not sure if I opened up the right file. So, let me just see. Yeah, looks like I looks like I did there. So, let's say you got small Android app, and then you can make a small change. Let's see. Uh now you can add the ability to duplicate a note. So, here you And so, as it goes and does its change, you know, it's going to be, you know, working in the code itself and then using those MCP servers that is exposed

31:36 by the plugin to kind of get that functionality itself. And uh what's also cool is, you know, you can do things where, if you notice this uh file opened up, um you can instruct it to open up files as it it edits it, right? So, as it kind of goes through the list of changes, you know, you can start opening it up and uh kind of showing the work that that it's doing itself in in uh in Android Studio. So, couple options here, you know, depending on what flavor you want. I I do think

32:06 that the ACP integration is is cool cuz you get a little bit more direct integration. Um but then you kind of miss some nice features such as the debug mode, which which I personally like, um and kind of the better rendering of of the uh information itself. But, nice to have the the two options there.

32:29 And uh yeah, I guess there's the third option where you use the MCP servers, and then also in ACP uh use the uh integrated experience there. So, that means the cursor agent will be able to use the MCP server inside of the ACP AI chat, which is kind of like all of them combined together Cool. I'll pause there. Check the chat for any questions.

33:00 Yes, Alec, is there anything in particular? >> No, >> [sighs and gasps] >> nothing too thematic to like re-hit on, but I would be curious like after going through and answering those, is there anyone that feels like they want like, you know, something expanded on or anything that Sunny can show or speak to that would be helpful?

33:18 I saw Nate recently ask is one faster than the other. Seems like agent mode was working faster, but maybe it was just the model. >> Uh by one, I assume you mean agent mode. Um yeah, it's it's going to be more model dependent cuz the the modes itself isn't really for speed, but it's more for direction. So, like how how it would kind of interact with you. Um so, with models, too, right? Like different model providers might have downtime. So, like Anthropic

33:48 models have downtime all the time. Um so, if you find that that's the case, you can always switch models or you can use auto. Auto will actually take that into consideration of, you know, the availability of the model um for the one we choose for the task at hand. So, that's another option for that.

34:09 And then uh yes, Edward, is there any Is there anything to say about React Native? Um yeah, I didn't go over over React Native. React Native is actually a much better integration cuz you can run Expo, the Expo server. If you if you are using Expo with React Native, you can run the Expo server, uh which will do like hot reloading of any changes. Um which I doesn't really require this whole like build and and and kind of run step. So, it's a little bit more straightforward there, but um, yeah,

34:37 it's a good call out. Maybe we'll add that to this this workshop for for next time. Cool. Well, that's all I had for you guys for mobile development. Um, if there's any questions on web development or, you know, just cursor in general for uh, any particular workflow, happy happy to field those as well since we have some some time here.

35:06 >> And I think in addition to that, is you also just have like general feedback to share like as it relates to like auto or any of the different features that we would love to understand that as well. >> I see we tour is there something regarding is there a new debug feature so we can check the screens. Um, yeah, for sure. So, you can actually instruct uh, through a skill for debug mode to have it uh, use those MTP tools as it runs the debug process. So, you can you can play around

35:48 with that. You could if you're on iOS, you can ask it to use the uh, review screen tool um, to see if that can kind of help uh, with that with its hypotheses that that it's working on with. Dimensions for multi-platform Flutter development. Um, I unfortunately do not.

36:08 I'm I'm not too familiar with with Flutter development. So, I don't want to uh, misguide you in any way. Um, but uh, yeah, I would I would look into kind of a similar setup if you're using a different IDE for Flutter development. So, uh, that ACP integration or kind of bringing up MCP servers to to expose functionality.

36:28 >> [clears throat] >> Uh James, um do you have any instructions that allow me to to get these MCPs hooked up? Uh I don't know if you have anything. Maybe I can write something up and and and share that out. Um what I can do though is I'll share uh that plugin that I'm using for IntelliJ, which has uh the Android uh MCP servers that it's using here. I'll add that to the chat there.

37:03 Um so you can you can use that to see what uh MCP servers are hooked up there. Um and then kind of like how the skills uh you know, utilize the tools that that it exposes. >> Uh I think that just Oh, you got it. You got it. Okay, cool. Um check if there's anything for freezing this via JetBrains AI plugin directly into Android Studio. >> Yeah, it is it's it's nice cuz it's it's integrated into that chat. Um I would say there's some I I I personally prefer

37:36 to do kind of like the sidecar mechanism with the CLI or the agent window. And uh it's just there's some like intricacies with how JetBrains implemented the ACP uh kind of client protocol side of stuff where like say if you're working on a plan and it builds a plan, right? There's not going to be like a build plan button. So you have to switch to agent and then tell it to build the plan. Um that's like one example of like a kind of oddity of of how how uh JetBrains implemented ACP. So cuz of that, I just kind of choose to do

38:06 that sidecar that sidecar option. >> And then so then one in the Q&A chat, Richard said, "We develop for an embedded Linux platform. We use Dart {slash} Flutter, and how could we connect Cursor directly to the platform for debug functionality? Currently, we have limited success through SSH." >> Yeah, interesting. Um I'm not sure. I'm not sure for sure. I'm not I'm not familiar with embedded uh Canvas platform or or Dart or Flutter. Um but uh curious uh for SSH through the agent window, um we have updated that uh to be

38:49 a little bit more functional for remote development than the IDE. So, you get you could try that out. Um yeah, maybe you could up if you get I see you're typing. Um what is what does that look like? Like are you are you running like a uh Flutter simulator somewhere else that is is running the embedded code? Um that could maybe help uh me me answer your question there.

39:13 And then uh I do see the other one, Florent, can you give more details about running Android emulators with Cursor Cloud agents in the network today? Uh not not yet. Um well, all right. You you could technically uh do that cuz Android is Linux-based, so um someone on the team did get it by a lot of instructions getting Cloud agent to run an Android simulator. Uh but we're going to be rolling out uh it's not rolled out yet. Um I don't have exact kind of date for that, but we are

39:42 uh working on a more integrated experience for that uh kind of across the board where you don't have to instruct the agent a lot to bring up the Android simulator. So, more on that in the in the future there. >> Awesome. >> Any other questions Sunny can help answer? Or again, feedback you want to share openly that we can pass on to the team?

40:07 Uh looks like Arthur has one. He says, "I have trouble deciding a rule versus skill for the mobile projects. Like for example, iOS has build profiles like preview, development, prod. Would a project rule be better to hardcode these variants or would a series of skills be more effective? Like for example, one skill for preview, one skill for dev, etc.

40:25 >> Yeah, it's it's a it's a good question. I would I would actually, at least from my experience with this with mobile dev is do anchor more on the skills. I find that uh putting in the rule for, you know, using a certain build profile and things like that can work, but having a skill is a definitive workflow that the agent, you know, will definitely run through versus it being more guidance for, you know, as it works. So, I find skills can be a lot more

40:55 effective for that. So, I would I would try that one out. >> And then Martin says, "I'm curious about policies. Like how do you design systems so that what gets built and shipped still respects Apple App Store and Google Play Store rules around code transparency and user safety? The MCP for research for Apple, the example about documentation, how far does it link to the required API native permissions for our apps within these policies? I'm assuming there's still manual work here to be done."

41:22 >> Yes, it's it's a great question. Um I'm not I'm not too sure fully on like the documentation piece for Apple and how much that can help kind of guide Apple review process and then Google Play Store process process review process. But, I think what could be cool there is, you know, creating a plugin where like a set of skills and rules that can you know guide systems to be kind of following those policies itself.

41:53 I think that would be a really good uh choice to have almost a skill like a set of skills that instruct the agent, you know, to go and check it as code gets developed. But, there's nothing out of the box for that. So, that's definitely something on the more customization end.

42:14 And then what about like skills versus commands? Like suggesting Yeah, I think you got that one. >> Yeah. Um Yeah, so uh commands have to be commands are uh manually invoked by you, right? So um they're very they're very similar, but uh skills have the option to be dynamically called by the agent, right? So the way skills are kind of uh loaded or given to the agent is through something called progressive disclosure.

42:39 So when the agent first boots up, right? It's going to have a reference to every name and description of all the skills you have. And as it's working, it can choose, you know, an appropriate time to pull in that skill and use that skill. And when it does that, it's going to pull in the rest of that skill.md file, right? So uh that dynamic uh kind of operating uh procedure of it, you know, will invoke skills when it's needed versus you having to explicitly, you know, do a

43:07 slash and the command to to invoke the command. Cool. And um any recommendation using advanced sensor-based features? Um I haven't I haven't explored too much, so I don't I don't have a recommendation. I I I assume you mean like sensors from from the device itself. Um What what could be interesting with that again is is uh extending capabilities through skills um and uh just thinking off the cuff here, like even if you're able to kind of like mock sensor data um through uh invocation of a skill, that

43:50 could be kind of interesting, but um no no recommendation off the bat here. Cool. Cool, just about just over 10 minutes left before the top of the hour. Anything else pressing while we while we have you all? If not, we can give you some time back. Awesome. It looks like maybe two more in the Q&A.

44:29 >> Uh so this is how can we container with applications? Um not sure exactly what what that uh what that means, but if you can maybe clarify um what do you mean by uh containers? And then uh Caleb, the description for the summit said it was going to cover how subagents could paralyze work across iOS and Android targets simultaneously.

44:50 Was hoping to see more of that. Yeah, good call. Um we'll definitely include that in the next session, for sure. And then I see there's one more from Ravi. Can this create the icon for mobile app? Um yeah, it's actually a cool feature that uh may may many people miss is that if you use Gemini 3.1 Pro, uh you can ask it to generate an image and then uh from your prompt, it'll create an image prompt and then from there you could generate icons or any other assets

45:21 you may need. >> Cool. >> Sounds good. Um yeah, if there's no more questions or anything, um we can maybe end a little bit early here. >> Yeah, for sure. This was this was a great time. There was a lot of good questions. Um I'll take some of the the Flutter and and kind of embedded Linux stuff back back to the team to see to see what they kind of uh recommend there, but yeah, I appreciate all you joining and being really really engaged. This was this was a lot of fun and uh

46:01 yeah, I'll see you next time.