Transcript

Be a 10x Vibe Coder (Claude Code + Cursor + MCP)

From Greg Isenberg · Episode page · Watch on YouTube

· 34 min · 6,911 words · 44 timestamped sections

This is the full, timestamped transcript of "Be a 10x Vibe Coder (Claude Code + Cursor + MCP)", an episode of Greg Isenberg published December 1, 2025 (34 min, about 6,911 words). It mentions 65 people, works and ideas, including Cursor, Claude, plan mode, Claude Code, anything.

What this episode is about

“Be a 10x Vibe Coder (Claude Code + Cursor + MCP)” breaks down a battle-tested AI coding workflow from Chris, a solo builder behind multiple revenue-generating productivity apps, showing how he combines Claude Code and Cursor side-by-side to ship faster. The episode gets practical on model selection (including when to reach for Opus 4.1), how to use MCPs correctly, what servers to run them on, and small but high-leverage tactics—like a Cursor hack for ~20% more output and a single keyword that makes Claude “work harder.” If you want a repeatable setup that turns AI tools into a real force multiplier (not just demos), this is a dense, actionable playbook.

Key takeaways

  • Use Claude Code + Cursor together—swap tools like gears, not like religions.
  • Save Opus 4.1 for gnarly problems; it’s elite but burns weekly limits fast.
  • When Opus is capped, Cursor’s Plan Mode is the “complex tasks” fallback.
  • Run Claude in a side terminal inside Cursor for a tight, always-on workflow.
  • AI isn’t a nice-to-have—it's how one person ships “impossible” apps solo.

Who speaks in this episode?

What is discussed, and when

  1. – 5:00: Cursor, Claude, Claude Code, AI agents
  2. – 10:00: Cursor, plan mode, AI agents, GPT 5 .1 high
  3. – 15:00: anything, Cursor, plan mode, AI calorie tracking app
  4. – 20:00: Cursor, Claude, plan mode, Claude Code
  5. – 25:00: Claude Code, Cursor, Claude, plan mode
  6. – 30:00: Cursor, Supabase, Context 7, Firebase
  7. – 33:50: deep research, BugBot, Claude, Cursor

Notable quotes from "Be a 10x Vibe Coder (Claude Code + Cursor + MCP)"

  • I brought on one of the best vibe coders on the planet to teach you how to squeeze the most out of Claude, Cursor, and all these vibe coding tools.
    about Cursor · at
  • There are two tools that I'm currently using. And again, I don't want this to be an episode to just kind of show how to use Cursor or how to use Cloud Code or anything, because…
    about anything · at
  • Is on the bottom, and you can just drag it to the right. And then I just run Claude inside on the right here. So now Claude is running. And then if I want to use cursor and…
    about AI agents · at
  • For very complex problems, I'm using Claude Code with Opus. And then, but if I'm not allowed to use it because I've exhausted the limit, what I'm doing for very complex…
    about plan mode · at
  • Choose between agent mode and plan mode and cursor. Plan mode just means that it's going to actually plan out its steps and thinking and you'll approve it. And here's the hot…
    about GPT 5 .1 high · at
  • I got to ask, because the Vibe Coding mobile app space is the next big thing, maybe. There's definitely a lot of opportunity there. A lot of people want to be able to do it.…
    about Rourke · at
  • This is just a way better way to do it. Second tip, this is specifically with Cloud Code. It's to use the keyword UltraThink. So this one is actually a special keyword that…
    about UltraThink · at
  • Really more digestible for Cloud Coder Cursor, and it will just use it way better than you trying to feed it in through a URL or something. So that's the first one. Second one…
    about Supabase · at

Who and what is mentioned in this episode?

People

  • Sam Altman — Person · 1 mention · first at
    For very complex problems, I'm using Claude Code with Opus. And then, but if I'm not allowed to use it because I've exhausted the limit, what I'm doing for very complex…
  • Alex Chris — Person · 1 mention
  • Bob Greg — Person · 1 mention

Companies & orgs

  • Cursor — Company · 32 mentions · first at
  • Claude — Company · 13 mentions · first at
    I brought on one of the best vibe coders on the planet to teach you how to squeeze the most out of Claude, Cursor, and all these vibe coding tools.
  • Claude Code — Company · 10 mentions · first at
    To run them on, and a hack that gets you 20 % more output from Cursor. There's even a single keyword that you can type that literally makes Claude code work harder. I'd never…
  • anything — Company · 6 mentions · first at
  • Supabase — Company · 3 mentions · first at
  • Rourke — Company · 2 mentions · first at
  • Context 7 — Company · 2 mentions · first at
    The hype was about with MCP servers. I now understand where they're useful, especially for coding. So there's two MCP servers that I'm currently using. The first one is called…
  • PostHog — Company · 2 mentions · first at
    Version of a lot of different documentation. And it's very compressed and very well formatted for an LLM. So instead of what I typically would do is I would just get the…
  • Firebase — Company · 2 mentions · first at
    Is, I'm probably going to get some heat for this, but because a lot of developers are probably going to say, oh, this is not secure. Why are you doing this? But in my…
  • createanything .com — Company · 1 mention · first at
    Any coding experience, people are always asking me, what should I use? Should I use Clawd code? Should I use Cursor? To be honest, my recommendation is to use, don't use these…
  • Vibe Code app — Company · 1 mention · first at
    I got to ask, because the Vibe Coding mobile app space is the next big thing, maybe. There's definitely a lot of opportunity there. A lot of people want to be able to do it.…
  • Google LLC — Company · 1 mention · first at
    Found 10 sources and then just put three circles to the left of it and maybe just find some random websites or just put, I don't know, YouTube, New York times, Google, just so…
  • GitHub — Company · 1 mention · first at
    Right now, there are a couple of AI coding tools that can review your code for you. And specifically, if you open a pull request in GitHub, so let's say you make a bunch of…
  • Amazon (links requirement) — Company · 1 mention
  • Anthropic — Company · 1 mention
  • Apple Inc. — Company · 1 mention
  • AWS — Company · 1 mention
  • Bolt — Company · 1 mention
  • Figma — Company · 1 mention
  • LinkedIn — Company · 1 mention
  • React — Company · 1 mention
  • V0 — Company · 1 mention
  • Xcode — Company · 1 mention
  • YouTube (as a source in UI list) — Company · 1 mention

Products & tech

  • BugBot — Product · 2 mentions · first at
    If you're a solo developer, you don't have that luxury. There's really no one who can review your code. So I'm using tools like BugBot, and CloudCode has one as well, where it…
  • Apple Notes — Product · 1 mention · first at
    So something that I kind of went semi -viral in my opinion for was the interaction or the animation for my daily or my calorie tracking app, Amy. When you search in Amy, it's…
  • Whisperflow — Product · 1 mention · first at
    Yeah, my AI coding has dramatically improved because I'm dictating everything now. So that's another hack. 100 % recommend dictating. You can use Whisperflow like I'm using.…
  • Amy — Product · 1 mention

Concepts & topics

  • plan mode — Concept · 13 mentions · first at
  • AI agents — Concept · 3 mentions · first at
  • UltraThink — Concept · 3 mentions · first at
  • terminal — Concept · 2 mentions · first at
    Show you, this is kind of the setup. So this is what it looks like for me. So I have cursor open. So this is cursor. And typically what I do is I have a terminal on the right.…
  • Opus 4 .1 — Concept · 2 mentions · first at
    That I'm using. It's Claude Code and cursor. In terms of the actual strengths and weaknesses of these, this is why I'm using it. So Claude Code is extremely good. I'm actually…
  • GPT 5 .1 high — Concept · 2 mentions · first at
  • Sonnet 4 .7 — Concept · 2 mentions · first at
    To plan out steps, maybe a writing model is actually better than a coding model at doing that. And then for the execution, when I'm done planning, I do use Sonnet 4.7, I think.…
  • security rules — Concept · 2 mentions · first at
    It actually might be more secure, because it has caught things that I have manually configured wrong. It has set up the security rules better than I would have done, to be…
  • production — Concept · 2 mentions · first at
    Such a powerful way. Just give Cloud Code or Cursor access to these tools. And anything you can do in the UI, they'll be able to do. When I'm in production, though, I will say…
  • pull request — Concept · 2 mentions · first at
    Right now, there are a couple of AI coding tools that can review your code for you. And specifically, if you open a pull request in GitHub, so let's say you make a bunch of…
  • deep research — Concept · 2 mentions · first at
    Other tools. But Whisperflow is really great because it has built -in developer terminology. So you can say developer terms and it kind of understands it already too. The last…
  • agent mode — Concept · 1 mention · first at
    Can start for free idea browser.com. And if you're listening to this, I highly recommend it. And I'm using and this is what plan mode looks like. You can choose between agent…
  • GPT 5 codecs — Concept · 1 mention · first at
    The hot take I'm going to have here. I am using GPT 5.1 high for plan mode. And typically, I'm seeing people using Sonnet or they're using even GPT 5 codecs. But for some weird…
  • AI calorie tracking app — Concept · 1 mention · first at
    Yeah. This one, yeah. This one fully. Fully working. The AI actually works. This one's like an AI calorie tracking app that I wanted to prototype. You take photos of your food…
  • Hello World — Concept · 1 mention · first at
    That, we'll just do some rapid fire tips on how I do this. So I have a blank Xcode project. This is just the starting Xcode project. When you run the app, it's just a simple…
  • simulator — Concept · 1 mention · first at
    If you're running the simulator, you can actually click at the top. It says slow animations. It will run it. I think half speed or slower. So you can actually see what's going…
  • slow animations — Concept · 1 mention · first at
    If you're running the simulator, you can actually click at the top. It says slow animations. It will run it. I think half speed or slower. So you can actually see what's going…
  • background tasks — Concept · 1 mention · first at
    It will work too. But there's no nice color animation or color thing here to show that it is actually doing something. So that's my second tip. The next tip is in Cloud Code…
  • server logs — Concept · 1 mention · first at
    This is a relatively new thing. But what this does is Cloud Code will literally run your server in the background. So it's running just like you would in the terminal. But the…
  • MCP servers — Concept · 1 mention · first at
    Server in the background for me. And you'll see this thing that says one background task, and it's basically running it. So I'm constantly using this. This is another big one.…
  • indexes — Concept · 1 mention · first at
    Hey, I set up the Supabase server, I set it all up, I hooked it up. Can you just double check that I set it up correctly? Check all the security rules, check all this stuff…
  • dictation — Concept · 1 mention · first at
    Worried about security, best $40 you'll probably ever spend to be able to sleep at night knowing everything is secure. The next thing is dictation. This is, as you saw, when I…
  • Codex — Concept · 1 mention
  • Favicon — Concept · 1 mention
  • iOS — Concept · 1 mention
  • MRR — Concept · 1 mention
  • prompting — Concept · 1 mention

Places & events

Sponsors

  • idea browser .com — Sponsor · 2 mentions · first at
    You think so too. Now, I think that you can look at trends to basically figure out what are the startup ideas you should be building. So that's exactly why I built idea…

Full transcript with timestamps

I brought on one of the best vibe coders on the planet to teach you how to squeeze the most out of Claude, Cursor, and all these vibe coding tools. We're talking about which models to pick, how to use MCPs properly, the exact servers to run them on, and a hack that gets you 20 % more output from Cursor. There's even a single keyword that you can type that literally makes Claude code work harder. I'd never heard of it. If you want to get the most out of vibe coding platforms, this is the closest thing to multiplying yourself. The full breakdown is inside this video, and there's just, it's a lot of tangible tips and tricks how to get the most out of vibe coding.

Chris, welcome back to the pod. By the end of this episode, what are people going to learn? Well, first, thanks so much for having me, Greg. What I hope people learn is just what my AI coding workflow is, because I know a lot of people are getting into AI, and there's a thousand tools out there. There's a thousand different ways to do this, and I want to show them what I'm using to build my apps. Then I want to also show some tips and tricks that I have that will probably help level up anyone who is doing AI coding. Correct me if I'm wrong, but you're doing thousands of dollars of MRR with your apps.

You've got a portfolio of successful apps, some beautiful looking things. So you're going to basically show us tips and tricks and your workflow so that people can copy it and do a similar thing? Yeah, yeah, exactly. Yeah, that's what we're going to do. All right, let's get into it. Okay, cool. So I'll just give a little bit of context for anyone who doesn't know me. My name is Chris. I build productivity apps, and then I document everything on YouTube. So I teach people how to build apps and just show what it actually looks like behind the scenes of doing it myself. I have a bunch of different apps that I've built, and the reason I'm showing them, it's not to plug them. It's really to show that these are very substantial apps. Some of them, or most of them, people are always surprised when they hear that it's just me building them. And to be honest, this would not be physically possible for a single person to do. If it wasn't for AI. So that

is my secret to being able to build all of these things and just sustain all this stuff myself, and just to leverage AI coding tools like Cloud Coding Cursor, like we're going to talk about, to be able to do this. So this is a little bit of context here. Let's just jump into it. So there are two tools that I'm currently using. And again, I don't want this to be an episode to just kind of show how to use Cursor or how to use Cloud Code or anything, because there are so many better videos out there. And there's a bunch on my YouTube channel as well, if you want to check them out. But I did just want to show what I'm doing and then some rapid fire quick tips so that you guys can maybe apply this stuff to your own coding workflow. But there are two tools that I'm currently using, and it's Cloud Code and Cursor. And what's surprising to most people is I am actually using both at the same time. And so usually people are like, I just have to choose one,

and then I got to stick with it. But what I found is they have different strengths and weaknesses. And that's why I choose to use both. And I'll actually show you, this is kind of the setup. So this is what it looks like for me. So I have cursor open. So this is cursor. And typically what I do is I have a terminal on the right. So typically the terminal I think is on the bottom, and you can just drag it to the right. And then I just run Claude inside on the right here. So now Claude is running. And then if I want to use cursor and their AI agents, I just have to change tabs. And then now here I'm using cursor. So I'm frequently switching between both Claude Code and cursor all the time for different purposes. And we'll get into why that is. But this is how I'm setting it up. And this is actually also how I do the iOS projects as well. Because a lot of people are always asking, how are you doing this with Xcode? Because I am using Xcode to do

the iOS projects. What I'm doing is I'm just opening the folder and the files in cursor. And now when I make changes here, it'll automatically reflect in Xcode. And then I can run the app and everything. So these are the current two tools that I'm using. It's Claude Code and cursor. In terms of the actual strengths and weaknesses of these, this is why I'm using it. So Claude Code is extremely good. I'm actually using this model called Opus 4.1. This is my favorite model. It's probably the most powerful model, the best model there is for coding. But it is super limited. They changed their pricing since Claude Code came out. It used to be unlimited usage. Now it's very limited to the point where I probably can exhaust the Opus limit for the week in maybe three or four hours, maybe less actually. So I try to use it very sparingly only for very, complex problems. So for very complex problems, I'm using Claude Code with Opus. And then, but if I'm

not allowed to use it because I've exhausted the limit, what I'm doing for very complex problems is I'm using cursor specifically with plan mode. Sam Altman, the co -founder of OpenAI, just said that it is the era of the idea guy and he is not wrong. I think that right now is an incredible time to be building a startup. And if you listen to this podcast, chances are you think so too. Now, I think that you can look at trends to basically figure out what are the startup ideas you should be building. So that's exactly why I built idea browser.com. Every single day, you're going to get a free startup idea in your inbox and it's all backed by high quality data trends. How we do it, people always ask. We use AI agents to go and search what are people looking for and what are they screaming for in terms of products that you should be building. And then we hand it on a silver platter for you to go check out. We do have a few paid plans that,

you know, take it to the next level, give you more ideas, give you more AI agents and more almost like a chat GVT for ideas with it. But you can start for free idea browser.com. And if you're listening to this, I highly recommend it. And I'm using and this is what plan mode looks like. You can choose between agent mode and plan mode and cursor. Plan mode just means that it's going to actually plan out its steps and thinking and you'll approve it. And here's the hot take I'm going to have here. I am using GPT 5.1 high for plan mode. And typically, I'm seeing people using Sonnet or they're using even GPT 5 codecs. But for some weird reason, and this is a tip I got from a developer friend of mine, he was like, you need to try GPT 5.1 high with planning. It's just for some reason so good. My hypothesis is that it was not technically built for coding, but it's really good for writing.

And when you're doing planning and doing a lot of critical thinking and you're trying to plan out steps, maybe a writing model is actually better than a coding model at doing that. And then for the execution, when I'm done planning, I do use Sonnet 4.7, I think. So I use the Sonnet model to actually execute. So that's the configuration that I have running right now. It's I use cursor plan mode. That's the important one. It has to be plan mode with GPT 5.1 high, Sonnet for the execution. And then if it can't get it, I use Clawd 4 .1 Opus, which again, I don't have much of that. So I try not to do it. And then I just switch between Clawd code Sonnet 4.7, so not Opus model, and then cursor plan mode. I just switch between them depending on the task. And then the reason, and here's when I would use which one as well.

So if you're using, if you're doing something very, very complex, like really complex, like it's a bug, I actually think Clawd, or I actually think Cursor with plan mode is better. If you're doing something massive, like you're trying to architect a whole app in like three or four prompts, Clawd is actually really good at doing that. And then if it's just really small details, you can get away with either, if it's not like something complex. And then I just switch depending on what it is. And then for some reason, Clawd is just better at UI stuff. So this is the configuration that I have going. And I have tried Codex, I've tried, I've tried almost all the coding tools. Right now at this moment of recording, these are currently the best ones. The other thing that I'll throw in here is, if you don't have any coding experience, people are always asking me, what should I use? Should I use Clawd code? Should I use Cursor? To be honest, my recommendation is to use, don't

use these platforms. It's actually to really start with something like createanything.com, for example. You can also get away using V0 or Bolt or something else. But what I found, if you want to do mobile apps specifically, I've tested all of them. Createanything is probably the best one at mobile specific. It's just really good at design. It's really good at following instructions. So my recommendation is to use a platform like that first. And then if you hit the limits, go graduate to something like Cloud Code and Cursor. And by that point, you should have a better understanding of how to work with AI and how to prompt it to actually use these tools. But this is the tool set. This is what I recommend. This is what I'm actually using.

I got to ask, because the Vibe Coding mobile app space is the next big thing, maybe. There's definitely a lot of opportunity there. A lot of people want to be able to do it. Rourke and there's the Vibe Code app are two popular ones in addition to anything. Have you seen them? Have you played with them? Any thoughts? Yeah, I honestly, I played with them. I still think, at least at the time of recording, anything is a bit better. And the reason is, for some reason, the design sense is a little bit better. And its ability to follow instructions is, it's just, it's so good.

It's so good compared to the other platforms. But again, you can get away with using any of them. They're all using the same models under the hood. I believe they're all probably using the Cloud models under the hood. It's really just also down to preference, I think. So if you prefer the style choice that Vibe Code or Rourke has, for sure, you can totally do it. But then if you, at least for me, subjectively, I did kind of like the designs that I was getting with anything. Fair, fair enough. I can actually show you, too. This is an example of, here was something that I actually did create in anything to show kind of what you can do. So this was something I fully designed in anything with the custom illustrations and interactions. And this does surprise people when I tell them that something like this, and again, these are like custom interactions that are happening here. Like these aren't

components or anything. All of this stuff was built in anything in, I think, less than 24 hours. Wait, that was built. This was designed and built in 24 hours? Yeah. This one, yeah. This one fully. Fully working. The AI actually works. This one's like an AI calorie tracking app that I wanted to prototype. You take photos of your food and it'll just immediately scan and figure out the calories and everything. Fully built. This one was using anything. 24 hours. Cool. So that's my recommendation on the tooling. And then in terms of models, we kind of went over this. We could go into a quick example of just kind of what it looks like, like what it actually looks like to vibe code something. So I know in another episode that we talked about, we were looking at interactions and animations.

So something that I kind of went semi -viral in my opinion for was the interaction or the animation for my daily or my calorie tracking app, Amy. When you search in Amy, it's kind of like typing in Apple notes. And then on the right, the AI is going to calculate all the calories for you and it'll just appear there. But there's this really nice animation that I have here where you can see that it kind of drops down. There's going to be this little shiny effect that's going on and it shows the sources. So what I wanted to do is kind of show you guys what it looks like, like how I actually code this stuff really quickly. So we can kind of go into that. And then after that, we'll just do some rapid fire tips on how I do this. So I have a blank Xcode project. This is just the starting Xcode project. When you run the app, it's just a simple Hello World app that should build. So we see it's a Hello World app. And again, I have this open in cursor and cloud

code. So what I'm going to do is I'm actually going to let's see if how far or let's just for a little bit, we don't have to do the full thing here, but let's try to replicate what I have here where let's just replicate the animation part where it does the searching, the calculating, all that stuff. Let's try to do that so you guys can see what this looks like. So the way that I would start is I actually let's just use cloud. You can again, you can either use cursor plan mode or cloud. Let's just use cloud for this case. And one of the tips that I have is I dictate everything just I get way more detailed prompts this way. And I'm using a tool called whisper flow as well to do this. So what I'm going to do is I'm going to say, can you modify this so that there's a button and when I click it, I want an animation to run. It's to simulate an AI searching for nutrition info and calories. So first I want

when you click it, I want it to say searching and I want it to have this shimmering text animation. Then after one second, I want that to drop down and then I want the text to say analyzing and I want that to drop from the top and replace it and it's going to take its place. And then it's still with the shimmering effect. Then after that, I want it to say found 10 sources and then just put three circles to the left of it and maybe just find some random websites or just put, I don't know, YouTube, New York times, Google, just so we can get the fabric cons for those, make those the sources on the left. And then after show calculating, then after that show some sort of hard coded calories, all just to show the animation. We just need to hard code everything. And then make sure that when the replacement happens during the animation, maybe we can make it so that it is cut off. So it's all contained

on one line. This is the level of detail that I go with my prompts. And then I just ask CloudCo to do this. So again, that's why I dictate because it would have taken me like five minutes to type this all by hand. And this is, this is kind of how I, this is how I scaffold apps. This is how I'm making my edits. It's just dictating everything. And I do the same thing in cursor plan mode. I'll just show you two. If we do the same thing in cursor plan mode, we won't run it there, but I can copy and paste what I just, dictated here. I'll run it in plan mode in parallel so we can see what this looks like. Okay. So now CloudCode, we're back in CloudCode and it's saying, I'll help you make the sequence. This will include this.

Do you want to do it? So it's asking permission if it can change the files. I'm just going to say yes. Go ahead and do it. So now it's running, it's editing files. And then we have cursor with plan mode. It's starting to plan out its steps. Again, we're not going to use this one. I just want to show you what plan mode looks like because it's pretty interesting. Okay. And then CloudCode is done. So it says that it did it. So let's just run it and let's just see what happens. Hopefully this is okay. Okay, nice. So, okay.

So it did actually make this animation, which is interesting. Obviously not to the level of polish that we need here. So I might make some quick modifications. So I'm looking really closely. And here's a pro tip when you're doing animations and interaction stuff. If you're running the simulator, you can actually click at the top. It says slow animations. It will run it. I think half speed or slower. So you can actually see what's going on here. So I can see here. Okay. Now there's even something going on when I do that. Like I don't like how it's dropping down and then going back up, it looks like. So what I'll tell it is something that's wrong with the animation is when something is leaving focus, it's shooting back up. Can you make sure that it drops down? Also increase the time to two seconds for each of the animations so they can show. Let's also make the shimmering effect a little more extreme. So we'll let this run. And then

here's cursor with plan mode doing the same thing. What I love about cursor with plan mode and why I think it's better at very complex problems is it's really good at following up. It's really good at asking questions. So here it's asking, should the button be usable multiple times? So we'll just say, no, we just want to rewrite. Actually, yeah, replay on every tap. For the three sources, are you okay with stabilized placeholders? Yeah, sure. Let's do that. So it's asking permissions on like what it should do. And then we have cloud code still running here. Okay, so now this is done. So let's see how much closer it got. And again, I don't think we're going to get as close as we want, because I think it took me like 20 prompts to get to the level I needed to go.

But okay, you can see it's a little bit better. It's already looking better already. Yeah. And then again, if you just do this a few more times, it's inevitable you can get to the level that I have here. That's exactly what I did here. I think it did take, again, 10 to 20 prompts to be able to get it here. But that's like an hour or two of time to get it to this level. And it's such a big difference compared to if we don't have it. Here's what it looked like before I added it. Again, there's no like it just kind of just says calculating. There's still kind of a shimmer effect. And then and then it shows the calories. But it makes such a big difference. So yeah, that's some that's that's currently how I'm doing it. Let's just go back to plan mode. So I'll just show you what this looks like. So when you're using cursor with plan mode, it will actually build out this plan for you. And so it's actually saying what it's going to do. It says step one, it's going to

do this step two, it's going to wait two seconds, say analyzing, and then you can review this plan. And then when you're ready, you can build it. And then here I chose 4.5 sonnet as the model to build it. And then when you click build, it's actually going to go ahead and do it. So that's that's actually actually I'm kind of curious what we can do is I'm going to completely restore this. So I'm basically just undoing everything Claude did. Let's just let cursor go ahead and build it. And let's just see what it comes up with.

So we can compare to see kind of what the differences between the two are. And then maybe you'll kind of see why I switch between them so frequently and why I'm not just sticking with one. I love having them sometimes do the exact same task and then just comparing and seeing which one is better. But that's who do you think? Who do you think is going to win? I, I actually think Claude code will win because I found that when it comes to design and interactions and animations, it does have a little bit of an edge, even if you're using the sonnet model and cursor for some weird reason, I think it is. Let's see if I'm wrong, though. Let's see. I'm actually curious. While we wait for this to go, though, what I'll do is let me, let me give you a few more.

We can just kind of hop into some tips. I know this one takes a little bit of time. So again, these are the tools I'm swapping between the two. These are the models I'm using. Opus 4.1, GPT -5.hi, and then Sonnet for the actual execution. Here are the tips that I have that really, I think, like when I've shared these tips, people are like, they, they usually pick something up from here where they, they add it to their workflow and they're like, this actually made a difference. So the first one, oh, actually, you know, this is done. Let's just see what's going on here. Okay, so cursor's done. Let's see what it was able to do. That's actually not bad. First try.

Actually, you know what, if we're comparing just the first shot, because again, that original first shot for Claude was a little bit more choppy. This one was actually really good. You can tell, though, it's so fascinating that it kind of came to the same conclusion. I think it's because, again, we're using the same model here. When I'm using Claude Code, it's using Sonnet 4.5. When I'm using the model here, I'm using Sonnet 4.5 for the execution. So that's why they look so, so, so similar. But it is actually interesting that we got a slightly better result when we did the plan mode first with cursor. So that's actually, I think I do, I would choose this one over the other one. I think, objectively, this one was better than the other one. First prompt, first prompt. I think so, too. I think so, too. This is a prime example of even if you're using the exact same model, if you just use plan mode, for example, versus no plan

mode on Claude Code, you can get pretty different results. That's a big reason why I think you should try to actually test it with a bunch of different tools, maybe have two AI tools that you're using for different scenarios. Yeah, this is a very fascinating one. To be honest, the rest of my workflow, this is kind of what I do. I just dictate everything. I'm constantly trying things. Let me go into some rapid -fire tips on how to get the most out of the AI coding tools. This is exactly what I'm doing in my workflow. And then if you need a basic tutorial on cursor and Claude Code, there's so many great ones out there.

The first one is, as we saw, there is a difference when you use plan mode. 100%. Every action that you take, use plan mode. It's almost like you give the AI time to think, and you will get better results. I already showed you what it looks like in cursor when you do plan mode. All you have to do is toggle plan mode here, and it's going to give this beautiful plan on the left. You can review it, ask for changes, and then when you're ready, you click build. Unfortunately, it's not as beautiful in Claude Code. So let me show you what this looks like. You can actually hit shift tab, and it'll change it to plan mode here. And now when you ask Claude Code to do something in plan mode, it's going to kind of do the same thing that cursor does, where it's going to think through, come up with a plan, and ask you to review it. But what I have found, at least at the time of recording, cursor's plan mode is stronger than Claude Code, which is why I use cursor

plan mode for insanely complex problems. It just somehow thinks about it a little bit better. It asks better follow -up questions. But that is tip number one. And then here's Claude Code asking for things. It's like, yeah, here's my plan. You can already just tell, like, oh, actually, this is not fair, because it's looking at the current code right now. But typically, this is what happens. It tells you its plan, and then it's up to you to just review and approve it. It will not be as detailed as cursor's plan. Cursor's plans are just so detailed. They're very, very good.

So that's tip number one. It's used plan mode. I'm pretty sure if you do this, the output will probably at least increase by 20 % at minimum if you just do this. So 100 % do this. It's also good to just review what it's about to do as well. It's better than just letting it run through your code base, and then you have to retroactively go back and try to fix it. This is just a way better way to do it. Second tip, this is specifically with Cloud Code. It's to use the keyword UltraThink. So this one is actually a special keyword that Cloud Code has. And I'll show you, if you type in, can you fix this issue, and you type in UltraThink, it actually changes colors, because it is doing something. When you type this in, Cloud Code is actually going to think a little bit harder about the problem. I'm assuming that it uses up more tokens and more usage when you do this. But to be honest, I think I use UltraThink in 90 % of my messages. And I

have not really found that it's impacted my token usage versus if I don't do it. It seems like it's about the same. So there's no reason not to do it. So when you're stuck on a complex problem, just type UltraThink. And Cloud will think, maybe, I'm assuming, it seems like it takes twice as long. So maybe it thinks twice as hard or something. But this is a keyword I'm very surprised people don't know about. So you will get a lot better results when you do this. I don't think there is one in cursor. It doesn't highlight. But I think you can still say UltraThink in cursor. If you're using Sonnet as the model, I think it will work too. But there's no nice color animation or color thing here to show that it is actually doing something. So that's my second tip. The next tip is in Cloud Code, you can actually run background tasks. So this is where you can tell Cloud Code, can you run the server in the background

for me? This is something I think they only added in the last month. So this is a relatively new thing. But what this does is Cloud Code will literally run your server in the background. So it's running just like you would in the terminal. But the important part is it now has access to your server logs. So when you're running and when you're coding with Cloud Code, it can now access this during debugging. So if you're like, hey, why is this not working? You don't have to copy and paste things and say, hey, the server's crashing, take a look at this. It just has access to it. And this is a massive one, which I just don't see people doing either. I think it's probably because it's a recent thing.

But I'm constantly telling it, hey, all you have to do is say, run the server in the background for me. And you'll see this thing that says one background task, and it's basically running it. So I'm constantly using this. This is another big one. The next one is MCP servers. So last time I came on the podcast, I was not using MCP servers. And to be honest, I didn't understand what all the hype was about with MCP servers. I now understand where they're useful, especially for coding. So there's two MCP servers that I'm currently using. The first one is called Context 7. It's a free MCP server. Basically, it just lets you access the latest version of a lot of different documentation. And it's very compressed and very well formatted for an LLM. So instead of what I typically would do is I would just get the documentation for something. Let's say I wanted to integrate PostHog analytics into my

app. I would typically say, by the way, here's the link to PostHog. Go implement it for me. But the problem with that is now it has to go scrape it. Maybe it doesn't do it correctly. Or you can use Context 7, hook up the MCP server, and you're basically giving Cloud, Code, and Cursor tools. And now it has access to Context 7, which now I can just say, oh, make sure to use the latest documentation for PostHog. Please use the Context 7 MCP. And when you do this, it's going to use Context 7, pull the latest documentation, really compressed, really, really more digestible for Cloud Coder Cursor, and it will just use it way better than you trying to feed it in through a URL or something. So that's the first one. Second one I'm using now is Supabase. So this one is, I'm probably going to get some heat for this, but because a lot of developers are probably going to say, oh, this is not secure. Why are you doing

this? But in my experience, this MCP, and you can use this for Firebase, you can use it for whatever database you're using, is so powerful to correctly set up a project for you. So I'm constantly using this to set up projects from scratch, because before I would have Cloud Code do all the coding and everything, but then I would have to manually go set up my database, make sure I'm configuring rules correctly. I would argue that if you are using a tool like Supabase MCP, where Cloud Code has full access to Supabase, it can spin up everything for you.

It actually might be more secure, because it has caught things that I have manually configured wrong. It has set up the security rules better than I would have done, to be honest. And so this is something where I actually do recommend, if you're a beginner or if you're a seasoned developer, it's really helpful to use this. And if you're a developer and you don't trust it, at least use the Supabase MCP to read your configuration and just double check your work. So you can just say, hey, I set up the Supabase server, I set it all up, I hooked it up. Can you just double check that I set it up correctly? Check all the security rules, check all this stuff, check the indexes so the performance is good. This is a really powerful MCP. And again, you can do this for Firebase, you can do this for AWS. They have an MCP, I believe, as well. And it's just such a powerful way. Just give Cloud Code or Cursor access to these tools. And

anything you can do in the UI, they'll be able to do. When I'm in production, though, I will say I'm way more cautious about using it, because it could just delete your entire server if it wanted to, and just delete the database. So I am a little bit more cautious, but I still do run this in production, though, because it's just so much faster for my workflow. So those are the only two that I'm seriously using right now. I'm sure that's going to change in a few months, though. I'm going to probably play with the Figma MCP now and test this out more. But yeah, these two are very good. The next one is, this is a very big one for solo developers or anyone that has no coding experience.

Right now, there are a couple of AI coding tools that can review your code for you. And specifically, if you open a pull request in GitHub, so let's say you make a bunch of changes, and you open up, if you have any coding experience, you know that the proper way to do something is you add your code to GitHub, and then another developer reviews it before it gets merged into production. If you're a solo developer, you don't have that luxury. There's really no one who can review your code. So I'm using tools like BugBot, and CloudCode has one as well, where it can actually live inside of your code base. And anytime you open up a pull request, or you want someone to review your code, it will do it for you. And I will say, these tools, when you hook them up, it is so much better than trying to do this inside of Cursor or CloudCode itself.

Because I think these are maybe custom models or something, but they're specifically trained to identify security issues and bugs. And so what I'm doing is, I have them hooked up to all my repos. And it has caught so many things for me. And again, it's stuff that I am a pretty experienced developer, I believe. But I mean, I'm working so fast, I'm moving so quickly, it's just inevitable that bugs will be introduced, security vulnerabilities will be introduced. It gives me so much peace of mind. And I think the Cursor one is just $40 extra per month. Totally worth it. Totally worth it for peace of mind, especially from the security perspective. So I recommend hooking this up. It's a hack for solo developers. And for vibe coders who are really worried about security, best $40 you'll probably ever spend to be able to sleep at night knowing everything is secure. The next thing is dictation. This is, as you saw, when I was dictating, when I

was trying to use cloud code, I was dictating my prompt. It's because you can just get way more detailed prompts. And then that will help steer the AI in a way better direction. I actually didn't think this was a big deal. But then when people started seeing me do this in my videos, I didn't know people weren't doing this. And that was something where people were like, wow, I just downloaded this. Yeah, my AI coding has dramatically improved because I'm dictating everything now. So that's another hack. 100 % recommend dictating. You can use Whisperflow like I'm using. There's also some other tools. But Whisperflow is really great because it has built -in developer terminology. So you can say developer terms and it kind of understands it already too. The last tip that I have is to use cloud for deep research on technical issues. So something a lot of people don't know is that if you are paying for cloud code, you also have increased usage inside of cloud desktop, like

the chat app. And you can actually use it to do deep research for you. And so what I'm doing frequently is I'm actually asking it, hey, this is what I'm about to do in my app. So again, I'm a pretty experienced developer, I think. So I ask it, what's the best way to store data? I'm looking for something similar to this in React. What do you think? And it will go and it will look for 12 minutes straight and look through all of the documentation and try to find the best way to do this. And then it'll just summarize everything for me. And then I can just take that and then review it, obviously. And then I'll send it over to cloud code and say, hey, can we actually do it this way? Because I think this is a better way. So this is a great way if you're a developer to double check, are you implementing them correctly, learn some new concepts. And then if you're just getting started out, this is also a

great way if you can just ask it, hey, I've never done this before. What is the best way for me to store data? So this is something that I am very surprised. I don't see a lot of people taking advantage of, especially non -technical people, especially people who are just starting to vibe code. You don't have to call your developer friends anymore and say, hey, how do I set up this? How should I structure my app? Just use cloud, especially if you're paying for cloud code already. And then obviously double check it. And if you don't know what you're doing, you can ask a follow up question. So you can say, why did you choose this pattern? Okay, what other patterns exist? There's also a great way to learn as well. So this is probably, it's a win -win if you use something like this.

And then you can just feed it into cursor and cloud code. And then it has access to this. It's able to implement it for you. And yeah, I've just, I found so many things that I've done not wrong, but I could have done better because I use Claude's deep research. So I know this wasn't a full on coding tutorial, but these were just a couple of things that I've been telling people in my YouTube videos. And I've gotten a lot of comments saying, yeah, these are very helpful tips. Like people did not know BugBot existed. People did not know deep research existed. So hopefully there's some cool stuff that you can take and apply to your own AI coding workflow. I appreciate you. I think we're all trying to get the most out of these vibe coding platforms and little tips and tricks help like this, especially a lot of us are, you know, solo builders, solo developers, right?

So it's not like we have people in real life to talk to. So this is helpful. And I just want to thank you for coming on and spreading the sauce with us. And I'll include where you can go find Chris and get more of him in the show notes. So you can check that out there. And Chris, don't forget about us. Come back on again. For sure. For sure. Yeah. Thanks so much for having me, Greg. Really appreciate it. All right.

Questions about this episode

Who speaks in "Be a 10x Vibe Coder (Claude Code + Cursor + MCP)"?

Greg Isenberg (Host), Chris (Guest).

What is "Be a 10x Vibe Coder (Claude Code + Cursor + MCP)" about?

“Be a 10x Vibe Coder (Claude Code + Cursor + MCP)” breaks down a battle-tested AI coding workflow from Chris, a solo builder behind multiple revenue-generating productivity apps, showing how he combines Claude Code and Cursor side-by-side to ship faster. The episode gets practical on model selection (including when to reach for Opus 4.1), how to use MCPs correctly, what servers to run them on,…

What is mentioned in this Greg Isenberg episode?

65 people, works and ideas are mentioned, including Cursor, Claude, plan mode, Claude Code, anything, AI agents. Each one on this page links to its own page with every episode that mentions it.

How long is "Be a 10x Vibe Coder (Claude Code + Cursor + MCP)"?

The episode runs 34 min; the transcript is about 6,911 words in 44 timestamped sections.

More from Greg Isenberg

About this transcript: the episode's spoken words as captioned on YouTube, split into timestamped sections. Reclip.Me locates every person, work and idea in the text, keeps the time it is first named, and quotes the words around it verbatim. Last updated December 13, 2025.