Page 1 of 1

MCP preview - my experience

Posted: Fri Nov 06, 2026 4:32 pm
by alan_sh
I've installed it and set the options.

I tried ChatGPT first, but as I only have a free account there, that didn't work (MCP only works with paid chatgpt accounts).

Pablo advised me to try Google Antigravity. So, I did. I set up the json file (downloaded it) and installed the Antigravity tool. It was all working fine until I ran out of credits! I now have a half build web site inside WWB preview. :)

So, the principle works, but I need to wait for next week and try something simpler.

Alan

Re: MCP preview - my experience

Posted: Fri Nov 06, 2026 4:42 pm
by AliGW
I got about four things done before I ran out of credits with Antivravity! Another AI bot told me that my system isnโ€™t really powerful enough to run it properly, so although it was fun while it lasted, as one would say in the den, โ€œIโ€™m out.โ€

Re: MCP preview - my experience

Posted: Fri Nov 06, 2026 5:45 pm
by Pablo
The limits for free accounts change frequently. When I started testing a few months ago, it was almost impossible to run out of credits. However, several powerful new models have been released recently, and these tend to consume credits much faster.

If you're using a free account, I'd recommend choosing a more affordable model, such as Gemini 3.8 Flash Low, to make your credits last longer.

There are also many other AI agents with MCP support, including Grok, DeepSeek Harness, Kilo Code, OpenCode, and Goose. I haven't had the time to test these myself, so I can't say how well they work or how they compare.

Unfortunately, running AI models isn't free. You either need to run them locally, which generally requires reasonably powerful hardware and higher energy bills, or rely on the computing resources of an AI provider. Some providers offer free tiers, but these usually come with usage limits.

As the developer of WYSIWYG Web Builder, I have no control over the pricing, credit limits, or usage policies of AI providers. However, I believe using AI agents through MCP generally offers a more predictable and manageable way to control costs than paying directly for API usage.

Re: MCP preview - my experience

Posted: Fri Nov 06, 2026 8:27 pm
by brimanHTML
I have not tried MCP public preview yet but will try this weekend.

I would suggest anyone buying tokens to take a look at monthly plans if you decide that you will be using AI provider for projects.

Using Claude Pro monthly as now I'm using agentic coding daily and buying tokens would be costing me way to much.

Re: MCP preview - my experience

Posted: Sat Nov 07, 2026 5:16 am
by wwonderfull
Prompt engineering was for those who know the 50% and wanted the other 50% for the ai to aid and reach 100%. Those who actually have a decent knowledge on codes know how to use ai even with free version and do not even use the paid ones.

If the website was very big I am 100% sure even ai token could not help maintaining such huge load of website with as less than hundred or even thousands of pages with text image and scripts. The automation of such would consume both Credits at the Same time your CPU GPU and RAM for crafting it on your computer.

It was meant for simple landing pages or 3 to 7 page website with some basic stuffs. AI does not know if the website has php errors or js console errors unless it is checked inside the browser.

You can think of MCP is for users for whom easy mode was not enough. Finding a block or a button was like looking for raccoons in the wilderness although they are just near the house hiding..

So now they get this new feature where they will prompt 3 times like a wish to a Genie and when you run out it will sleep for weeks and demand credits before waking up. That a commercial Genie right there.

Re: MCP preview - my experience

Posted: Sat Nov 07, 2026 6:42 am
by AliGW
wwonderfull wrote: Sat Nov 07, 2026 5:16 amIf the website was very big I am 100% sure even ai token could not help maintaining such huge load of website with as less than hundred or even thousands of pages with text image and scripts. The automation of such would consume both Credits at the Same time your CPU GPU and RAM for crafting it on your computer.
I was working on an update of just one page. Granted, I was using an 'expensive' model, as I was unaware at that stage that there were different models to be chosen - everyone is new to this at some point, and there's a learning curve that some of you people seem to forget. But the reality is that my laptop just isn't powerful enough - for one thing, it has no dedicated GPU.

I have the free ChatGPT installed and can work for hours with it updating things, with it making suggestions and me implementing them. I will try a 'cheaper' model of Antigravity once I get access again next week and see how much further I can get, but I think the limitations of my system will be the sticking point, and I am cool with that. I can't afford a hardware update at the moment, nor can I afford a monthly subscription to AI. And guess what? I am cool with that, too.

I was not complaining, merely sharing my first experience. I am learning, as others will be, too - using AI in this way is completely new to me, but I am a quick learner, even at my age (I have an agile brain - I always have had), so I am catching up fast. :P

In other words, don't underestimate me.

Re: MCP preview - my experience

Posted: Sat Nov 07, 2026 7:25 am
by Pablo
It was meant for simple landing pages or 3 to 7 page website with some basic stuffs. AI does not know if the website has php errors or js console errors unless it is checked inside the browser.
MCP can do much more than you might think! It can inspect the generated HTML, CSS, and JavaScript, analyze the output, and check for potential errors. It can even inspect how your website looks in a browser, take screenshots, and simulate PHP execution, provided you give it permission to do so.

I think you're underestimating just how powerful this integration is. The AI has access to an enormous amount of information about WYSIWYG Web Builder, including features and settings you might not even know exist.

Sometimes it feels like it knows more about WYSIWYG Web Builder than I do ๐Ÿ˜‰

Re: MCP preview - my experience

Posted: Sat Nov 07, 2026 7:47 am
by AliGW
That's the impression I got from my very short interaction with it, Pablo. What it was able to do in that short time was really impressive.

Even the free ChatGPT can interrogate locally published files and spot potential issues. I don't doubt that an AI able to use the WWB MCP will make our websites much more efficient. I was blown away by what it could do.

I would caution anyone considering using the AbacusAI Bot that I tried the other day, though: I found that it took it upon itself to start 'fixing' files on my system that I had not even asked it to touch based on a theoretical question (yes, I had made it clear that nothing was to be done immediately). I would not risk using it again. It was able to reverse the changes it had made, fortunately!

Re: MCP preview - my experience

Posted: Sat Nov 07, 2026 8:04 am
by wwonderfull
Who needs ai credit when we have pablo and our forum members right. MCP is a good bonus feature I would say and is great to have. But we should still keep our skills sharp on wwb in case the call of caveman ever comes : )

Re: MCP preview - my experience

Posted: Sat Nov 07, 2026 2:55 pm
by brimanHTML
Prompt engineering was for those who know the 50% and wanted the other 50% for the ai to aid and reach 100%. Those who actually have a decent knowledge on codes know how to use ai even with free version and do not even use the paid ones.
I can understand both sides of pro AI and those that oppose AI for many reasons.

The reason for seeing Agentic AI as just prompt engineering is false to a degree. If you are vibe coding and just hoping for great results sure, valid. If you are someone that knows what you are coding and can prompt wisely then it is a different beast.

Not to drift to far off the ecosystem of web here but I recently updated a project I hand coded for work that is C, Python, Bash and was able to use AI to update many features in hours and also generate Unit Testing, very successful.

I can understand the push back on AI as I started coding for web back in the 90's, among other such as C and Pascal.

The best thing a developer can do is understand and adopt technology that can assist them, not replace them BUT assist them.

I was the same a few years ago but have seen how AI can assist, not replace but assist development and very happy in my current development environment with AI.