Transcription
New AI features are being added to VS Code every single day, just about. And I want to show you five. Yes, five. Okay, actually six if you stick around for a bonus one at the end. New features that I think you're absolutely going to love that change the way you use agent mode and code completions every single day inside of VS Code. So, let's get into it.
[Music]
All right, let's talk about MCP servers. First and foremost, it's easier to see what you have installed either inside your project or as a user inside of VS Code and also install them. If you go over to extensions, you can see, of course, your installed and recommended extensions. You'll also see a new MCP server section here. Now, I don't have any installed at a user level or inside of my project. So I can click on this MCP servers. This will bring me to the VS Code page where I can browse different MCP servers available. Now, there's of course tons available out there, but it's kind of a curated list of things like GitHub, Docs, Markdown, Hugging Face, Notion, Zapier, Memory, Firecrawl, Clarity, Stripe, Azure DevOps, Azure. Let's go ahead and install GitHub. So I want to have it at my root all the time. So I'm going to say install GitHub. It's going to open up VS Code and give me this nice page to explore the MCP server. So, I can actually browse all of it here, manifest, go to the repo, everything like that. It's kind of like an extension, right? And it even shows me how to set it up here. But I can just click install and boom, it's here, right? It's awesome. I can go ahead and at any time go into settings. I can say start server, show output, configuration, model access, and more. So, I'm going to say start server, and then it'll just log me into GitHub. Just like that. And now when I go to my tools, I can see all of them built-in and my GitHub MCP server. So I have everything authenticated just like that.
Now, this isn't creating a new MCP.json file in my VS Code folder. I can go up here and say MCP. And when I do that, I'll see list server, browse resources, reset cache tools, trust, show installed servers, list servers. Now I'm going to see the open workspace folder or open user configuration. The workspace, the one that I just talked about, the mcbp.json, but the open user configuration is one that's sort of in the application root for anytime I use VS Code anywhere, and that's where it's configured automatically. So, I can see it right there. Of course, if I don't want it there, I can just copy and paste this directly into a new MCP.json file. Anyways, MCB servers are easier to install than ever.
When I'm using agent mode, I'm often adding new features, functionality, fixing bugs, and tuning a lot more. And I often wanted to build a solution or projects and run tests. So when I do this normally, let's see what happens. So if I come in and say, how do I build this solution? Uh, what are the commands, right? So I'm just going to ask it what it is. And it's going to tell me here probably to build it. Do a dotnet restore, a dotnet build, a dotnet run. So I could say, okay, let's build it. So now what it's going to say is, okay, cool. And build. Let's run this command, which is something that I would do all the time, but it's going to ask me to continue. You don't really need to ask me to build a project if you're implementing stuff. Just go ahead and build it.
So now, if you go into settings and if you go ahead and go down to extensions and go down to GitHub Copilot under experimental, in this case, it may be in preview or automatically enabled by the time you watch this, but there is a new allow list and a deny list. So you can specify what commands you want it to automatically run without any prompting. So if I go in and say add .net build and hit okay here. Now we can go back and I could say, let's build it. And now it will automatically run that build. This is so nice when working with agent mode to specify what commands you want it to automatically run. It's super nice to go in, have your allow list and your deny list, anything that you want. So, PowerShell, bash, anything like that, specify them here. So, I have dotnet build, dotnet run, dotnet test, have everything automatically set up.
Okay, one of my other favorite features is a new setting. Just go into your settings and say max requests. All right, this is key. This defaults to 20. So, when you're working in agent mode, you often have it have long-running operations. The default for max request before it asks you to continue is 20. Set it higher. I set it to 100. Set it to 10,000. Just whatever you want. It'll just keep going and keep grinding until it finishes. And it'll never ask you for continue ever again. Combine that with the allow deny list and that is a game changer.
All right. All right, I showed this other one in another video, but I'm going to tell you again. Go into your settings of the agent mode. Hit generate instructions. This will automatically go through and analyze your entire codebase. It will create and or update your Copilot instructions, giving it a full overview of everything in your project, how to run it, how to add features, and a whole lot more. It will analyze your best practices, your coding conventions, and a lot more. So every time you make a call with agent mode, it will automatically send these, and then you can specify them even more details if you want to. As you make changes to your project, go back in, go inside of here, play around, hit generate instructions again. You can run it with different models, anything that you want. It's basically running a big prompt, and it is a game changer when working with Copilot instructions inside of VS Code or Visual Studio or anywhere that you're using Copilot.
Okay, I love code completions. When I'm typing code, which I still do, I say group dot and it automatically fills in this ghost text, this code completion recommendation that I can just tap, tap, tap, tap, tap and go, go, go. Sometimes you're in a flow, you might be demoing, you want to do something. You can now click this little Copilot icon down here and you can configure if you want code completions on for all files, the type of file you're in, if you want next edit suggestions, or what I like is this little snooze button. You can add five minutes at a time. So you can automatically snooze the code completion. So now if I come back over and I say group dot, I don't get any code completions coming in. So I can just write. I still get my IntelliSense. I still get everything that I want, but now I'm just back in the flow. I can come back in. I see there's a little Z right there. If I tap on it again, I can go ahead and hit cancel. Come back in. And now I'll start to get my code completions again coming in just like that, which is super awesome.
Okay, the last thing I want to talk about is the awesome Copilot repo on GitHub, or has a bunch of custom instructions and prompts and custom chat modes which basically let you identify behaviors for tools when working with agent mode. So things for example like planning mode or DBA or PRD type chat creation inside of here, refining requirements or issue chat, and you can basically scope down what tools are available and give it additional instructions when working in that mode. My favorite is this one from Burke Holland, 4.1 Beast Mode, which for all intents and purposes gives, when working with GPT-4.1, additional behaviors that do much more in-depth planning and execution. So it goes through and he worked really close looking at all the OpenAI documentation and identified this workflow and understanding what it needs to do. So let's go ahead and put this chat mode in. I'm going to go to RAW. I'm going to copy this. I'm going to go over into my build and I'm going to say settings modes and then I can add a new chat mode file in my GitHub chat modes here. And then I'm going to say Beast Mode. You can even add them like system-wide and paste that in.
Okay. So now I just have a single product endpoint here. Let me say, let's add a new user endpoint and create a new web UI in the front end for managing them. Now here I could use just normal agent mode, but I'm going to go into the agent mode custom Beast Mode. Now I can still select the different models, but I'm going to go ahead and select just 4.1. It's been optimized for and hit go. Now, what I like about this is that 4.1 is going to now identify all the steps that it needs. It's identified eight different steps that it needs to execute like data entities, P data context, the endpoints, the program files, and a lot more. And it's going to go and update along the way of everything that it is doing. So, this is really, really neat as it adds files. And this has been a complete game changer for me when I'm working with agent mode and 4.1 because it is super duper quick.
Those are just some of my new favorite features when working with agent mode and code completions inside of VS Code. But there are so many to explore. What are your favorite features? Let me know in the comments below. If you enjoyed this video and you've tried out some of the features, give it a thumbs up, share it with a friend, and don't forget to subscribe. Jam that notification bell so you get notified every time I put out new videos right here on YouTube. So until next time, I'm James. Thanks for watching.
[Music]