Cursor CLI Tutorial for Beginners (should you use it?)

Key Takeaways

Last week, Cursor released their version of clawed code, the cursor CLI. Now, this is a brand new tool. You can use it in any command line. It doesn't have to be inside of the cursor code editor, and it uses a lot of cursor's core technology, like the ability to read files, call commands, make changes, and review the code changes that it makes. It's a brand new tool, so it's not as feature complete as Clawed Code, but it does have certain features that make it stand out. So, let me show you how to get started and we'll build a large new feature with it. Also, make sure you stick around for the end of the video because I have some important considerations to share when it comes to using the CLI. First, to get started, you're going to want to run this command in a terminal. Now, again, it can be in any terminal. It doesn't have to be in cursor. I'm just using cursor because that's what I'm used to. Now, if you're on a Windows, you're going to want to use WSL and then you can run the command. Once you've run it, you may have an additional step here which you should follow to make sure you can call it in every future terminal session. Once you've done that, you can use cursor agent. Now, if it's your first time using it, it's going to ask you to authenticate. You can go to the URL, sign in with your cursor account, and you should be good to go. And then you will see a screen like this. And you can think of this in many ways as being the same as the cursor chat. So, you can pretty much tag files in here, ask it to do different things, pretty much all the things that you would do in a normal chat. And I'm going to go ahead and start by planning a feature. And I have a certain prompt that I've already written for this. You can check out my workflow video if you want to learn more about this. But essentially, I have this prompt that I like to use for planning features. Now, Claude Code has a dedicated custom slash commands where you can do something like this. But in cursor, you can get the same effect by just putting your prompt in a file and then tagging it. Okay, so I'm tagging this feature and let me describe what I want to build. Actually, before I do that, let me go ahead and switch the model because right now it's going to use GPT5, but from my experience, I found Sonnet 4 to be a better option. So, I'm going to go ahead and do model Sonnet 4. And now we're going to use Sonnet 4. Let me go ahead and plan feature again, and I'll describe it. Okay, I've written the description and I'm going to go ahead and submit it. But basically what I'm trying to do is create a Trello like board to keep track of new features for my product Vibe Scan because we launched it last week and we got number two product of the day on Product Hunt. It was super cool. Thank you all for your support and now we have a bunch of users and they have a lot of really cool ideas on how to improve the product. So I want to be transparent about the things that I'm prioritizing and I think this is going to be a great way to do it. So, let me go ahead and ask the cursor CLI to plan this feature. Okay, we can see that it's creating a to-do list for itself. And now it's starting to search and read files. So, it's doing a lot of the same thing that cursor would be doing in the chat or that clawed code would do. Okay, so it looks like it's just getting to work instead of planning the feature, but I really want it to create the plan. So, let me go ahead and clarify that. I'm going to reject this command and I'm going to give the follow-up message. I said that I want you to plan the feature before implementing it using guidance from the plan feature document. Okay, it wants to run this command to figure out the next feature number that it should use. I can go ahead and click run for that. Okay, there we go. Now it's actually created the plan. So I can scroll up here and I can actually control-click and it'll show up right here and I can see the contents of that file. So let me go ahead and review this. I'm leaving it some feedback, but one issue I have with this, which I also have with clawed code, is the fact that when I press shift enter, it doesn't go to a new line. I really hope these command line tools can fix that. I gave it some feedback. Now, it's going to go ahead and update the plan. Okay, looks like it's made changes to the plan. Let me take a look. Things look good to me. So, I'm going to go ahead and ask cursor to implement phase one. But before I do that, I'm just going to clear the context so that it's not confused by all the other random files that it was looking through before. To do that, you can press slashcle and it's going to start a new chat session. Here we go. Please implement phase one of and then I'll tag the plan. It looks like we're done with phase one and it has created five different files or edited them. And the cool thing about cursor CLI that I do like a lot is that they have the ability to review files. So I can click controlr and then I can actually see the changes that were made. Now this file is new and I can just use the arrow keys to navigate. Right now all of these files are pretty much just new except this one. So we see a couple of changes here but this will really come in handy as we change files and review the code in phase two. So, I'm going to click escape to go back to the chat. And now I'm actually going to go ahead and start a new conversation again. And I'm going to say, "Please implement phase 2A of my plan." And what's cool is now I can actually run these in parallel. So, I'm going to open up a separate command line and go into WSL cursor agent. And in here, this is a separate chat. I'll say, "Please implement phase 2B of the plan. So, I can have multiple of these agents running in separate terminals. Now, as it's running, I see that it's struggling to update this file and it's kind of inserting things in the wrong place. So, I'm just going to ask it to write the file from scratch. It looks like it messed up another one of my files and I had to do a get checkout to revert it to the proper state. So, it does seem like there is some difference in how it's applying these changes from the cursor agent because I've never seen this issue come up when just using the cursor chat. Okay, looks like it's finished the UI changes. I'm going to go ahead and let it build the UI. Okay, looks like it was able to build it. So, now I'm actually going to go ahead and clear this chat. And by the way, if you ever want to go back to a chat, you can type resume and then you'll see these different chat IDs. And when you exit a session, it'll also tell you the ID of that particular session. So you can go back to it. But I'm going to start a new chat because I'm going to say, "Please follow the code review instructions to review completion of feature 33." While it's running that, I'm just going to check it out in the UI. Okay, we see an ideas board now. I'm going to click on that. Looks like there is an API issue. So maybe the review will catch what the problem is. Yep, it looks like there's a data alignment issue that the review identified. So now I'm just going to go ahead and ask the reviewer to fix all of these problems. Okay, looks like it changed a bunch of files. I still saw some errors. So I went ahead and tried to build it and there were eight compilation errors. I'm actually going to copy these and then I'll just paste them in here. And we'll see it pasted like this. So you can actually also drag and drop images in here to attach them or or files like this. Now in my case, I just need the logs. So, please see compilation errors. It looks like it was able to fix those issues. And now when I go back, I well, I at least see something. Now, in light mode, that actually looks kind of nice. But we're going to want to figure out the styling when we're in dark mode. Let's try suggesting an idea. Let's add the ability to ignore findings. It's like if you run a scan and it finds some issue in your code that you know is something you're okay with, you want to just ignore it so you can focus on the issues that really matter. Let's see. Submitting an idea. All right, cool. Well, we have this here. Looks a little clunky. Um, I'm going to want to adjust it, but can I move it over? Nope. Something went wrong. I'm going to write down all of my feedback, open a new CLI chat, and then iterate on it, and I'll show you the final product. Okay, so after about an hour of work, I got it to look like this. As an admin, you can move these across the different columns. And then as a regular user, you can see those updates. And you can also suggest ideas on things you would like to be implemented. Let's add a feature to fix all issues. I'll submit it. It's pending review. Now, as an admin, I can see it. I can improve it and move it over. And there we go. The feature is complete. But there's a big elephant in the room that we need to talk about, and that is cost. Because I flipped over to the cursor dashboard, and it looks like I spent $30 on the CLI without even realizing it. This is a major problem because not only was it expensive, it was also not something I realized I was doing until I went to this dashboard. So, I really hope the cursor team at a minimum surfaces the cost as you're running the CLI so that you at least realize what is going on because it looks like it was using the max mode for a lot of these and with 5 million tokens in the max mode. Yeah, that becomes pretty expensive. So, after I did that, the rest of these requests I actually just ran using Sonnet 4 using the regular cursor agent because it looks like the cursor CLI just defaults to using the max mode. In fact, I'm not sure if it can even work without the max mode, and that isn't super clear when you're using it. Now, in terms of output quality, it worked fine. It was pretty fast. I did get better results out of the cursor agent inside of the cursor editor. I'm not sure if that's just luck or how I worded my prompts, but I ended up cleaning up a good amount of work in the main cursor editor after I had done the initial implementation in the CLI. But here's the thing with all these CLIs. I think they really shine when you use them inside of a workflow instead of this sort of interactive approach. And you can do that with the cursor agent by passing in an API key parameter and then you can run it as part of your own code that you write. And that's exactly what I did with claude code when I implemented the autofix feature inside of vibe scan. You can see exactly how I did it in this video. So I'll see you over there. Thanks for watching. Take care.