One click MCP Superpowers for Local AI 🤯 | OpenClaw, Claude and Hermes

Added: 2026-05-12

[00:00:06]Hey guys, welcome to the show. Today we're going to be bringing the power of MCP to allow us to create images and videos straight out of Open Claw and your favorite inferencing applications.

[00:00:16]So this right here is Hex-Rays MCP and it turns homes into creative engine, Claw, Perplexity, Open Claw, all this kind of stuff. And rather than having to figure out their interface on how to generate images and videos, as you've previously seen on this channel, you can just get your inferencing application to do it all for you. So there's a couple of ways to set it all up and have it all integrated into the Claw and your favorite harnesses. Now, this one supports Claw, Perplexity, Open Claw, and all that kind of stuff. So with Claw, you open up the connectors here and you copy in that custom link. With Perplexity and Open Claw and that stuff, you have this copy and paste link. So if I fire up the Claw, so I've got my Claw here connected up to Inferencer, my inferencing application. As you can see, we're loading up Qwen 3.6 the 35B1 cuz that one is super fast, it's MMLU. So it's making a tool call.

[00:01:02]See, it's generating 50 tokens a second there. Another 50 tokens a second here.

[00:01:06]This This window is very, very useful to find out exactly what's going on in the background while Claw is doing its business, what it's doing. So I'm not sure what Open Claw is doing. I think one of the issues I have here, I am running the 35 billion parameter model, and so it's probably not sophisticated enough to do that whole MCP connection at the base. So what you actually can do, and this is the method that I'd first do anyway, if you go into Inferencer and click on tools, you can just click add tool and connect the MCP server yourself. Now, the good thing about using it inside Inferencer sandbox is that it is a macOS sandbox application, which means that it only has access to resources that you give it access to. It's kernel level sandbox, that means it can't execute code on any of your folders and do some naughty stuff. Whereas other harnesses like Open Claw and all that stuff, it pretty much had access to your whole Anything that you can access on your user, so just be careful about that. So what I'm going to do is I'm going to paste in the MCP link and jump into Infront side just hit add and connect.

[00:02:01]And now I'm in here I'll click on the connect button. And it's waiting for access and it's giving me a link to connect to my MCP. So I'm just going to copy that and here it's saying, do you want Infront side to connect to your application? Be careful cuz you can find out your email address and your identity. Now the application itself doesn't actually lock this stuff. It just takes the token, but other applications could know this information so just be wary of that. So I'll click on allow and it's going to fire Infront side back up. And as you can see now we have all of these tools that we have access to. So we got models, explore, we got generate video, we got jobs display media upload, virality predictor. That one looks very interesting. Virality predictor predicts a video's virality, potential engagement, attention, all that kind of stuff. So I'm going to enable these tools and inside Infront side here, I'm going to also enable the tools here. I'm going to jump back into claw and I'm going to just going to ask it, what tools do you have access to? And look right there, it's saying it's got access to the image and video generation tools.

[00:02:57]So I'm also going to jump back into the claw and because I've got that tools button enabled and the claw is accessing the server that I'm hosting on Infront side and Infront side says allow Infront side tool calls. When I go into open claw here, I'm going to say, what tools do you have access to? It should now give me a mesh between the tools provided that we connected via MCP and Infront side as well as the tools that it has access to and it can run in its Freelancer environment. And look at that, it says Hicksfield MCP is connected. So even though we didn't connect Hicksfield directly via open claw, we connected it via Infront side and I'm not sure what the latest situation is with open claw, but I know previously they used to like save their tokens in public visibility. Within some we actually save it within the keychain of Mac OS so it's like a little bit more a lot more protected than what you get out of here. So look at that, it says okay, so what image models do you have access to? Now one thing I I don't like about Open Claw, is that I don't actually know what's going on. I can see that it's making like a million different back and forth requests.

[00:03:58]And it's it's doing something in the background to figure things out. So, I just prefer to just prompt the model directly inside the inferencing application. So, if I go here and I continue this conversation, so what image models do you have access to?

[00:04:11]You can see I'm just pretty much it makes a tool call, accesses the website, and look at that, all these parameters are returned by all the image models that it's requested via their MCP API. So, we know we can access Nano Banana, Flash, Clin, Sea Dream, Flux, all these kind of image generation models. So, I'm just going to say, "Can you generate me an image about about Open Claw?" So, it's giving me some options here, and this is I'm using the creative power of Clin 3.6 5.5B to brainstorm this situation. So, I'm going to say um creative illustration and Nano Banana to be creative with the prompt I want to put the image on a t-shirt.

[00:04:58]So, it's now calling the MCP generate image tool, and the prompt is a bold edgy t-shirt design featuring stylized mechanical claw emerging from digital glitch effects. So, you see it says come back and it says, "Perfect.

[00:05:12]I've submitted your t-shirt design."

[00:05:14]And the image is being generated. Can you update me on the generation? So, it's MCP job status. I'm just talking to it, finding out what's going on. It says the image is still being generated.

[00:05:24]Here's the current process. Two two seconds before next turn. Can you check again? Your image is ready. And it's giving me a preview URL. And that is the image that we just made. So, now that we've got the image, I'm going to do something even cooler. I'm going to say, "Can you make a video from this image?"

[00:05:42]So, I'm going to see what it's going to do. What is it going to do? It's using Sea Dance 2.0, a dramatic scene where a mechanical claw emerges from a t-shirt design and boom, the job has been submitted. So, I'm going to ask it, "Show me my latest generations." So, I should hopefully see the image that we just made as well as the video we just made. We just got a big chunk of text.

[00:06:03]Now, obviously, I don't understand what that is, but the large language model is trying to figure it out. 8,000 tokens, this one's very, very fast. But, boom, so we made the latest video is the mechanical claw video. I want to watch that video, so I'm going to click on that.

[00:06:23]>> [laughter] >> Yeah, it's a good video. Go see that straight up. And the image we just made should be that one.

[00:06:38]Boom, and that was a t-shirt that we just made uh previous videos. Now, we can also do this stuff in Claw. So, I'm just going to ask it the exact same question, "Can you list me the latest generations?" And I'll also at the same time I'm going to ask it over at Infrasound, "Can you viral rate that latest video?" And then we got our results back from Open Claw, so we can see that we also got some videos, images, videos. Now, unfortunately, you can't click on his list. Let's say, "Can you show me the last two?"

[00:07:11]Hopefully, I want to see the images and videos just like we clicked them on Infrasound. So, Open Claw again, did the same tool as it did last time, except now it's actually got the links there, so I can click on that and view and just opens it up in the web browser as a link. So, nonetheless, that is still good. So, we got Open Claw connected to the MCP server. And we also got the virality predictor running over here, all being done on MCP. And the cool thing about it is uh this stuff even Infrasound, it's got an iOS application, so you can do all this kind of like stuff when you're out and about on the go. You don't even need to go on the website if you want to do some large language modeling. And what's cool is you can get to kind of like research or maybe set a cron job, especially open close really good at that. It's got cron building which is a timer, so you can say every morning read the news and generate a video based on that news or all the kind of cool stuff. So, let me know what you guys think of MCP coming to open close, Hicksfield, and Interseller. Hope you guys found this video useful and enjoyed the show.

[00:08:14]Oh.