0:02 Claude Code has a content creation
0:04 problem and Higsfield's brand new MCP
0:07 server just solved it. Instead of having
0:09 to individually connect every AI content
0:11 creation tool to Claude Code, which you
0:12 kind of have to do since the best ones
0:15 change from week to week, we can now get
0:17 them all in one place via the MCP
0:19 server. And this just isn't a
0:21 convenience win. This means we can now
0:24 reliably automate large portions of our
0:26 content creation process with the best
0:28 AI tool for the job. So today I'm going
0:30 to show you exactly how to install this
0:33 tool and the process I use to create
0:35 this piece of content that got 100,000
0:38 views in less than 24 hours using the
0:40 Higsfield MCP server. So why does the
0:42 Higsfield MCP server even matter? Why
0:44 should you care? Well, I alluded to it
0:46 in the intro, and there's two reasons.
0:50 First one is the fact that we now have a
0:53 single pathway to getting access to
0:56 every single AI content creation tool.
0:57 Because up until now, we really haven't
0:59 been able to do that programmatically.
1:02 Instead, you had to individually connect
1:05 all these tools to claude. Giant pain in
1:07 the butt. Nobody did this because
1:09 everyone had its own API, its own
1:12 payments, even if the API was publicly
1:14 available, which in some cases it
1:16 wasn't. And so, you were kind of locked
1:18 into the one or two you actually used.
1:21 Problem with that is the best ones
1:23 change all the time. Last week, Dano
1:25 Banana Pro was on top. Guess what? Now
1:29 it's GPT images too. 6 months ago VO3
1:31 was the top dog. A month ago it was
1:33 Cling and today it's Seed Dance. Are you
1:36 using the best tool for the job? Chances
1:37 are you weren't if you were set up like
1:41 this. But now alls I have to do is be
1:43 connected to the Higsfield MCP. And
1:47 boom, I can now connect to all of these
1:49 and more. And by more, I mean there's 17
1:52 image models, 14 video models, and we
1:55 have access to a lot of the Higsfield
1:57 proprietary models. But the real unlock
1:59 isn't the convenience. It's the fact
2:01 that since it's an MCP server, we can
2:03 now automate a lot of these processes
2:06 via Cloud Code. For example, I can
2:08 create an automation where every single
2:10 day Claude Code takes a look at GitHub
2:12 and it says, "Hey, what are the top
2:16 trending AI repos for this week, for
2:18 this month, brand new ones that just got
2:19 released?" It's going to take that
2:21 information. It's going to bring it back
2:23 inside of Cloud Code, and it's now going
2:24 to structure it in a way that I could
2:26 use for some sort of social media posts.
2:29 In our example, it will be a carousel.
2:31 Alongside that, it will then create a
2:35 prompt for the images so that we can get
2:38 images like these, but it has to copy
2:40 and the information from the GitHub that
2:41 it just pulled. It's then going to send
2:43 all that information to Higsfield, which
2:46 will then call on GPT images 2 to create
2:48 all that for us. It then brings it back
2:52 into Cloud Code. And voila, we just have
2:53 a completely automated content creation
2:55 process. From there, I can manually
2:57 review them. I can have Claude Code post
3:00 them. But the point is I can now
3:02 automate some sort of flow. You could
3:04 automate even more of it where I'm
3:06 grabbing information from some outside
3:08 place in this case GitHub. I'm then
3:10 analyzing the information inside of
3:12 Cloud Code. I'm taking that analyzed
3:14 information and I'm turning it into some
3:15 sort of content prompt which gets sent
3:17 to the Higsfield MCP. And then it brings
3:19 it all back to me and I have a nice
3:20 deliverable and I really haven't even
3:22 lifted a finger. So that's the real
3:25 power that is unlocked via this MCP
3:28 server. So you combine these two things
3:29 and we are really turning Claude Code
3:31 into a marketing machine. So let's talk
3:33 about the install. First of all, you are
3:35 going to need a Higsfield account. I
3:36 will have a link to that in the
3:37 description. If it wasn't clear by now,
3:40 Higsfield is a one-stop shop for all
3:43 things AI content creation related.
3:45 Next, we need to install the MCP. And
3:48 there's really two ways to do this. One,
3:50 we can go inside cloud.ai and just set
3:53 up a connector. Two, we can just do it
3:55 inside the terminal via Cloud Code.
3:57 Setting up the connector is very easy.
3:59 You'll just go to claw.ai. You'll head
4:01 to settings. You will go to connector.
4:05 You will go to add custom connector.
4:09 You'll copy this, paste that in there,
4:11 and hit add. You'll then hit connect.
4:13 And it will ask you to log in. And boom,
4:16 there we go. I can now essentially call
4:19 any of these audio, video tools, image
4:22 tools that live inside of Higsfield from
4:24 Claude itself, the web app. And I can
4:25 also do it from the desktop app. So
4:27 inside the chat I said use the Higsfield
4:29 connector and create an image talking
4:30 about the power of cloud code plus
4:33 Higsfield using GPT image 2. And you can
4:36 see it's calling the model right now. It
4:38 will ask you for some permissions. You
4:40 can see the actual prompt it's sending
4:42 in JSON. And we see the image in
4:44 progress. The nice thing about doing
4:47 this inside of the actual chatbot
4:49 application or on your desktop Cloud app
4:51 is the fact that the images will be
4:53 generated in line, which means I'm
4:54 actually going to be able to see them.
4:55 And remember, there's a lot more we can
4:58 do than just simply create an image or a
5:00 video. There's actually a lot going on
5:02 under the hood with this MCP. You can
5:04 ask Claude itself to explain it to you,
5:06 but I also have this guide that I wrote
5:07 up that I will put inside of the free
5:09 school community. There will be a link
5:10 to that in the description. And here's
5:13 the image it created for us. And as I
5:14 scroll over it, you can see I have a few
5:16 options. I can recreate it. So
5:18 essentially just send the prompt there
5:20 again. I can animate it. So send it to a
5:23 video editor. I can edit it. And so what
5:25 it does is it brings up essentially
5:27 another prompt. In this case, it would
5:29 send it to Nano Banana 2, but I could
5:31 change that to, you know, like GPT
5:35 image 2. It links the reference image so
5:36 it knows what it's editing. And then you
5:38 just put your prompt in there. So pretty
5:40 intuitive in terms of how you mess with
5:43 this inside of the chat application. But
5:44 let's talk about what I think is the
5:46 biggest unlock and that's using it
5:48 inside of Claude Code. So to set up the
5:51 MCP server inside of Claude Code, super
5:52 simple as well. Literally just going to
5:55 use plain language and say set up this
5:57 MCP server for me. You'll just head back
5:59 to this page which is the Higsfield MCP
6:02 page. I'll link that as well. You'll do
6:04 custom connector.
6:06 You'll paste it in there and it's going
6:07 to go to work. It's going to set it up
6:08 for you and it will also give you a link
6:10 to go through the same authentication
6:12 process you just saw me do on the web
6:14 app. Now to confirm it's set up just do
6:17 for/mcp and you should see Higsfield
6:19 connected. If you don't just have a back
6:21 and forth with cloud code and it will
6:22 walk you through the steps to make sure
6:23 it's connected. You may just need to
6:25 exit cloud code and spin it back up. Now
6:28 at this point once the MCP server is
6:31 connected we can now use basically any
6:34 AI content creation model from the
6:36 terminal through natural language. So if
6:40 I tell cloud code create me 16 different
6:44 images with GPT images too it will do
6:45 that for you and it will just download
6:47 them and you can even tell it hey I want
6:49 you to bring up the images for me. The
6:50 only downside with the terminal is
6:52 obviously we can't see the images inside
6:56 the terminal itself. But hey, what we
6:57 really want to do is we want to figure
6:58 out how to put this inside of an
7:00 automation, how to script it. But
7:02 natural language prompting is simple and
7:05 exact same process as I showed you on
7:06 the web app. So let's go through this
7:09 process. So what we first need is we
7:10 need to be able to grab information from
7:12 GitHub and bring it back into cloud
7:13 code. And you can see that right here.
7:15 This is an automation that runs every
7:17 single morning and it grabs the top 10
7:20 trending GitHub repos this week that
7:22 were created in the last 7 days and
7:24 ranks them based on stars. It gives me a
7:26 quick description all this sort of
7:28 stuff. And I can also see the top five
7:30 trending over the last month. And again,
7:32 these are just new ones that are just on
7:33 the scene. Now, to create this for
7:35 yourself is actually really simple. I
7:37 have the whole breakdown inside of Chase
7:39 AI plus, but you can literally just
7:40 prompt Cloud Code and say, "Hey, can you
7:43 create me an automation that checks
7:44 GitHub for this every single day?"
7:46 There's no API you need to set up or
7:47 anything like that. But what I want to
7:50 do is I want Claude Code to take a look
7:53 at this information and I want it to
7:55 essentially turn it into a carousel. And
7:56 if you're not familiar with carousels,
7:58 they're just posts like this. We'll have
8:00 some sort of cover page. This one is top
8:02 five Claude Code front-end skills, but
8:05 instead we'll do top five Claude Code
8:07 GitHub repos or top five AI repos. We'll
8:09 see what Claude Code comes up with. I'm
8:11 going to give it the reference images
8:12 that you see here. So, I'll give it the
8:14 cover page and I will give it some of
8:16 these, you know, body slides, so to
8:17 speak, because I'm going to want it to
8:20 be in the same sort of style. And we'll
8:22 see what it comes up with. So, I'll feed
8:25 it this. I'll feed it the GitHub
8:26 information. And then cloud code should
8:28 say, okay, based on everything in this
8:31 GitHub, based on the reference images,
8:33 here's what we should think about in
8:34 terms of coming up with a prompt. So I
8:36 gave Cloud Code a pretty simple prompt.
8:38 I said, take a look at our GitHub
8:40 trending data for today. What I just
8:41 showed you inside of Obsidian, I want to
8:43 create a carousel talking about that
8:46 information. We could call it top five
8:47 trending AI repos this month or
8:49 something like that. I want it turned
8:52 into slides like this, cover plus body
8:54 slides. And then I fed them those four
8:56 slides or those three slides. And then I
8:58 just said, "Let's talk about it before
9:00 sending it off for content creation."
9:03 Now, what we're doing here is we're sort
9:05 of manually going through each step. So,
9:08 we already did the content. Now, we're
9:09 going to talk about it before we send it
9:12 off here to Higsfield. What you would
9:14 actually want to do after you kind of
9:16 got this to a place you liked and you've
9:18 continued doing this over and over is
9:19 instead of me being, "All right, now
9:21 let's do GitHub. Now, let's talk about
9:23 it. Now, let's push the prompts. We
9:25 could actually turn this entire thing
9:28 into like one large call it you could
9:30 call it like Higsfield skill or really
9:32 any skill you want to give it whatever
9:34 name you want. But you can automate this
9:36 entire process and you could have
9:38 something that every single day, you
9:41 know, right after it hits GitHub in the
9:42 morning and says, "Hey, here's the top
9:44 10 repos." Well, why don't we just turn
9:46 that into a post? You could have a
9:47 carousel every single day that says,
9:50 "Hey, here are the top 10 trending AI
9:53 repos for today." You know, that's
9:55 actually like somewhat relevant content
9:56 that people would actually like. And
9:59 this is an easy way to create it.
10:01 Don't steal my idea. So, Claude is
10:02 telling us, "Hey, I pulled today's
10:05 trending files. It's just repeating the
10:09 top five GitHub repos for this month. It
10:11 has some thoughts. Claw code is
10:15 suspicious." Yeah. Yeah. Little sketch.
10:17 talks about the hook angle, talks about
10:19 the title
10:21 as well as the layout and the hero image
10:23 and all these things. So, here's the
10:24 prompt I gave it. It talked about using
10:25 a carousel skill, which is actually an
10:27 irrelevant skill for this. It has
10:29 nothing to do with Higsfield MCP. So, I
10:30 said, "I ignore that skill. Let's start
10:32 with the cover slide." So, that main
10:33 slide that everyone's going to see, I
10:35 want it done in the same style as a
10:36 reference image I gave you. Use your
10:40 copy. Use Higsfield's MCP. You use GPT
10:42 images, too, for variations.
10:44 And that's kind of wordy, which is
10:46 totally why you would eventually want to
10:47 turn this into an actual skill if this
10:49 is something you're doing a lot. So
10:51 remember, we are trying to create
10:53 something that looks like this because
10:54 we are feeding this exact reference
10:56 image in there and we're saying do
10:58 something similar. Just change up the
11:00 copy, change up the title. So it just
11:02 came back with the four variants. It
11:04 took about 5 minutes. Understand the
11:06 speed at which this is going to happen
11:08 is going to be totally dependent on the
11:11 model you use and the quality. So for
11:14 GPT image 2, I was doing high quality 2K
11:16 and I wanted four variations. Another
11:17 thing you need to think about when you
11:19 do this is the way the MCP works is you
11:21 were just sending a request. It's not
11:23 going to hit you back up when it's done.
11:25 So you need to tell Claude Code, hey, I
11:28 want you to pull Higsfield every 60
11:30 seconds, 90 seconds to see if it's done
11:32 and then bring it back to me. So here's
11:35 the four variations. We got one, two,
11:38 three, four. Now, we pretty much told it
11:41 do the exact same thing. Just put our
11:43 new copy on it. And it did exactly that.
11:45 Actually, I think it looks pretty good.
11:47 If I wanted to edit something, I'd
11:48 probably get rid of the list down here.
11:51 I'm not sure if I am am in love with the
11:54 Chase AI up top, but point being, if you
11:55 tell it, hey, use this reference image,
11:57 it's sending the reference images just
11:58 like if you were in there doing this
12:01 manually. And so, step one, giving it
12:03 some sort of reference image to use as a
12:05 cover, did a solid job. Now, let's see
12:06 how well it does here in terms of these
12:08 body slides. Now, you'll notice here
12:11 we're actually grabbing some stuff from
12:13 the GitHub page itself. So, what I'm
12:15 going to tell Cloud Code to do here is
12:18 find your own assets that we could use
12:20 as reference images that would make
12:21 sense for the value page. Again, like we
12:23 have the full power of Cloud Code here
12:26 to help improve the quality of the work.
12:28 So, I said first slide looks good. Let's
12:32 move into the body slide. Use the first
12:34 GitHub repo that's up. And then I said,
12:37 hey, go ahead and figure out what assets
12:38 we need from the GitHub itself to be
12:40 used in this generation. Research that
12:43 GitHub, pull assets as needed, add them
12:45 to the MCP request as well. So, I'm
12:46 having it do quite a bit here, like go
12:48 on the internet, find the appropriate
12:50 repo, grab what you need from there,
12:52 bring it in to your prompt, and then
12:53 push it to the MCP. And here's what it
12:55 came back with, and it was giving us the
12:58 slide for awesome.md
12:59 for reference. This is what the
13:04 awesomedesign.md GitHub looks like. So,
13:06 pretty close. I think that looks good.
13:09 And it gave us four variations, all that
13:11 are slightly different. Nothing really
13:14 popping out, but I think it did a really
13:16 good job. It also definitely matches the
13:18 aesthetic that we gave it here in the
13:21 reference image. So, really, really
13:23 good. And now all we would do is repeat
13:25 that exact same process for the rest of
13:26 the slides. And we wouldn't have to go
13:28 one by one at this point. We could
13:30 essentially have it rapid fire all of
13:33 them. And so you could see how easily we
13:35 can turn something like this into a
13:37 content machine, especially if we have
13:41 some resource like daily updated GitHub
13:43 repo list. This is just like an
13:45 evergreen content thing that people
13:47 would be interested in and I can do it
13:48 all from here. I can turn this into one
13:51 single skill with the MCP server really
13:53 powering the creative side. Now, the one
13:56 thing I will also mention is you don't
14:00 have to go full AI image generation for
14:01 all these things. Like you could also do
14:03 sort of a hybrid style where we use
14:05 Higsfield for the cover image because
14:07 this is where the aesthetics is really
14:09 going to play a big role. And then maybe
14:12 instead you want to go lower cost, lower
14:14 tokens and for the sort of body slides
14:16 you use like HTML or something. You have
14:18 cloud code essentially generate that via
14:20 code. Lots of ways to approach this, but
14:22 the big thing is is we have we have
14:24 options now. We have options now with
14:25 this MCP server. So, that's where I'm
14:27 going to leave you guys for today. All
14:28 the links for this stuff can be found
14:31 inside the the description. Make sure to
14:33 check out Chase AI Plus if you want to
14:34 get your hands on my Cloud Code
