Anthropic Just Dropped the Feature Everyone Asked For

Okay, so a few hours ago, Anthropic released a brand new feature, which many people are calling Anthropic's own version of OpenClaw, OpenClaw for grownups, and now saying they don't even need to use OpenClaw or ClaudeBot anymore because they can just replicate the same thing inside of Claude Code. And this new feature is being able to control Claude Code running locally on your machine from your phone or another device. Now, you could have kind of done this before by using tmux, Tailscale, and also Terminus on your phone... And I've covered that before in my Claude Code Masterclass. But that was like kind of a hassle to do. And this update makes it much simpler because now you can just use a Claude Code app, kind like a messaging app.

Now let's go for an example of how you can use this. Now you want to go to a folder. I'm going to go to my RayOS folder, which contains all my life stuff that I manage with Claude Code. And then do claude rc or claude remote-control. Press enter. And then that will spin up an instance of Claude Code. So the very first time you run it, you may see something like this where you have to enable it. And that opens a secure connection from your computer to Anthropic servers. Now I'm actually gonna run this in dangerously skip permissions instead. So let me stop that, do Claude dangerously skip permissions so it can do more things on my computer. And what I can do is do /remote control to enable it for this session. And if I want to enable it for every single session that I run in the future as well, I can do /config search for remote control. And it says enable remote control for all sessions. I'm going to change this to yes or true and then press Escape. And now remote control will be available for every session of Claude Code that I run in the future. So anyways, let's make sure that we have the Claude app installed, open it up, log into our account, then go to the code tab on the left-hand side and then find our session. Can you go to my Google Drive folder and then figure out where the raw MP4 file is and then edit the video with the, uh, video editor skill. So as soon as I send that message, you can see it now appears on my computer. So it's first going to go to my Google Drive, but I should have been a bit more specific because I have Google Drive set up locally. And recently I actually made a brand new skill for Claude Code to edit my videos for me. So this video that you're watching right now has been edited by Claude Code, but it's only like 95% perfect. So I still have to like clean up some loose ends myself. Um, but yeah, now you can see it's now downloading the video that I want edited. And whilst that's happening, I'm going to start a brand new session. So I could also be coding on projects, for example, like I have this project right over here and then I can remote control this session. Since I already had it enabled before, it's already enabled. So I can go back. Can you make the macOS release bump up the minor version? And yeah, that command sends off to my computer and it will work normally. Now the nice thing is that every time there is a permission that your computer requires because you're not in dangerously skip permissions, for example, you can allow it. And that is one of the benefits over using tmux and Terminus for this because it was like quite fiddly to allow it from your phone.

Now, one of the benefits about this is that you have access to all your Claude Code config from your phone. So whatever skills you have set up, MCP servers and so forth. So if I run claude rc, then I can trigger plan mode and also the Exa MCP server that I have enabled from my phone. So let me go to the session over here. Switch over to plan mode. And if I press enter, you can see it's now using the Exa MCP server that I have set up on my computer. Now this means that you can have MCP server set up for background research tasks, for example, and then you can just be triggering the use of them from your phone. And now we can see that planning mode is asking us some questions that we can enter in from our phone. So I can just say like, hello, press submit, and I can answer a lot of questions that way. So that can be pretty helpful for just making plans on the go because you can go out for lunch and then continue working on your application. And now going back here, we can see Claude Code is still editing that video for us. So let's talk about

the most important thing here, which is security. You don't want to be on bypass permissions or dangerously skip permissions all the time because that can get quite dangerous, especially if it's connected to internet. Now your version of Claude Code, which has remote control enabled, is connected securely via Anthropic's own servers to your phone or to another machine. So if I wanted to, I could go to the Claude AI website from another machine And then I can see that I have this remote session right now, which is the same session that is happening on my phone. And the server remote session is the one that is editing the video for us. But if I was kind of worried about it accidentally being prompt injected by searching online or deleting a file, I could either do one of two things, either use the Claude Code sandbox or have Claude Code running on a remote server somewhere. So I'll show you how

you can install Claude Code on a remote server so you can have it run dangerously skip permissions over there and do like web research for you and stuff like that. Without worrying about it accidentally deleting your files or doing unintended things on your computer. So I'm gonna log into my Hetzner account and then just quickly set up a server for myself here. So you can do create server, choose the cheapest one, which is $3. 49 a month. Choose— choose Germany over here, create my SSH key and add to that here. And then call the server Claude Code and press create and buy. And now what I can do is SSH into that server copy over the install command for Claude code, go back to the server, press enter. And then after it's installed, I can do Claude, press enter. But I actually have to copy over this command to fix the bash, do Claude, sign in with my Anthropic subscription. And then finally I can run tmux on the server, run Claude inside of tmux. Yes, I accept. Do /remote-control, enable remote control for the session... And then I can detach from this terminal session safely and have it running on the cloud. So if I press Ctrl+B and then type in colon detach, press Enter, then I can go back to my phone's version of Claude Code. And then I can see I have this interactive session and then I can chat with it and say stuff like searching online, tell me about Claude Code. So I could install my MCP servers onto this server and then have it continuously work for me in the background on different tasks and not have to worry about my computer being on all the time. So I can do research with different MCP servers, make reports for me, and even if it did get prompt injected, any damage that it does would be limited to the server that it's currently on. Now, another way of doing this is you can use

sandboxing in Claude Code and have your remote control sessions run through that sandbox. So I cover much more about sandboxing in my master Claude Code class that will be linked down below if you're interested. But to give you a brief overview, if you make a folder like this and inside of your settings. local JSON file, you type in something like this, sandbox enabled true, permissions default mode don't ask. I can give it a list of websites I can just go to and do research on. And then I can deny any tools such as editing or writing tools, for example, and bash tools. And if I start a remote session of Claude Code here by doing claude rc, I can go to a session which is in the sandbox research environment here and then say, go on archive and find me the latest AI papers and summarize 5 that look interesting to you. You can see it can go and archive and not ask me for permissions because that is allowed over here. And then after pulling in the papers, it gave me some interesting summaries of them. So if I told it to fetch masterclaudecode. com, then you can see that it won't work because that URL has been blocked. But one of the problems can be that it may still decide to go on the website via a curl command, which can be dangerous. But in this case, it does not have access to bash tools, so it won't be able to do that.

Now, one of the ways to make this stricter and more secure is to use a proxy. So for example, I would want it to be able to do curl requests, but I want to deny the WebFetch tool and the WebSearch tool because those would bypass the proxy. And then I should set up a network port over here. So that is HTTP proxy port. And then you want to make a proxy that looks kind of like this so you can get Claude Code to write you one. And you want to make sure that api. anthropic. com is allowed because that allows for the remote connection. Because if you don't, you may run into some problems. Now if I enable the proxy by doing node proxy. js, then do claude rc dangerously skip permissions. And then I can test out the proxy to make sure that it works by trying to go to my personal domain. So that should fail because, like, it's not allowed by the proxy. So you can see on the computer, it says HTTP GET blocked. But if I tell it to, like, do masterclaudecode. com instead, then that should work just fine because that is allowed by the proxy. So you can see, passed successfully via the proxy, and then it loaded in the page. So this is another way that you can make sure that you are using remote control securely, with dangerously skip permissions. So you could have like research agents that are only allowed to go on a set of websites, for example. You can also block a lot of permissions to prevent anything bad happening on your computer. And it's like really easy to customize.

Now, one of the big differences from OpenClaw is that it's more passive and less proactive. So you can get OpenClaw to like search online at a scheduled interval for you. And then send you messages. Here it's more like passive, so you give it and tell it to do something, it does it for a while. But given how popular it's gotten, I imagine that Claude Code or the Anthropic team will slowly be adding them over time because it seems they kind of have an interest in the space. One feature that I would really like to see is basically the ability to connect hooks to notifications on your phone. So currently you can have Claude Code send you notifications on your computer using like Apple Scripts or something, but if Anthropic allowed for those notifications to be sent to your phone if you're remote controlling, I think that would make it even more powerful.

But yeah, I'm sure they will be adding a lot more to it over time to improve it. Now it may be the case that you no longer need to use OpenClaw because this satisfies a lot of use cases for you. And I know for me personally, because I usually manage everything in my RayOS folder, the fact that I can now use this remotely from my phone means that I can just have it running all the time and then just be giving prompts to my RayOS folder to like draft emails, download files, do some video editing, and a bunch of other things as well, whilst using my Claude Code subscription to save on money. Now Claude Code has finished editing that video, so I'm going to get it to edit this video and then finally upload it. Anyways, I'm sure Anthropic I think will be making much more improvements here over time because they seem to be shipping a new update like almost every single day. And if you do want to get my own thoughts about it and ways in which I'm using it, then you can sign up to my newsletter, which will be linked down below. You do get access to a bunch of bonus free videos as well that you won't find here on the YouTube channel.