Todays status of the MCP

Share your work Show and Tell
1

Not sharing my work really, just wanted to create an appreciation post regarding todays state of MCP.
I can’t share any screenshots because of confidentiality, sorry.

I’ve been building with weweb for a few years now, and my project is quite big. It also has a lot of different quirks from the state of weweb at the point the different parts was built. Thus it is a mix of old and new, and its just a lot to handle.

In the start of my journey with weweb we had only a AI inside the formula editor. It was really good, and I could choose what variables to give as context, and it built javascript for me. I loved it, and it really helped me develop. Then, the new weweb AI came out, and I tested it a few times, over the iterations that came from weweb, and not a single time was able to get a usable result, even thought I tried to manage context as good as possible. Could not create a successfull page, could not fix a formula, nothing. Absolute misery.

Then this summer: Enter the new MCP. A new era.

I did not have much chance to test the mcp, just a few small things here and there. With success.
Today I had some time spare and I first created a Claude Skill to use the Claude Chrome editor to navigate the to the page to visually check it’s own work.

Then I gave it two quite big tasks:

First agent: Create a page for listing customers using my own AG grid component, handle access management, create backend functions in n8n.

Second agent: Create a page to edit customers, addresses, contact persons, and all connected to a customer. Handle access management, create backend functions in n8n.

At the end I asked it to use my weweb skill to visually check the work in the editor. Some back and forth with some adjustments, and voila. It basically oneshotted both my requests. 1,5 hours of agentic work and both was finished. Weweb pages, backend n8n flows, access management, all perfect.

I’m thorougly impressed with the state of the MCP. Some small things missing, like it could not add a custom component, and some workflows was hard to reach, but it just stopped, asked for help and voila.

Claude Fable 5.1 on high

Going forward I will create a skill for claude to understand how my project works so it does not have to spend too much time and context on understanding the setup. And I will create a weweb skill to put things we learn while working with the MCP, so we never do an error twice.

This is just great, keep up the good work WeWeb!

2

Very cool! Thanks for taking the time to share this @thomlov :slight_smile:

I love building with the MCP as well. Now have a folder where I’m slowly adding skills dedicated to building weweb projects (e.g. generating a pre-publish checklist after making changes) and general instructions on how I like my apps to be built (e.g. naming conventions, dedicated view in detail pages instead of frontend filters on all the rows in my table, etc.). Each weweb project now has a dedicated folder where I store the pre-publish checklists, product roadmap, claude design files, etc.

It’s still a work in progress but I love the fact that the work you put into this compounds and improves your building efficiency over time. It’s really exciting and empowering!

3

Very cool, I might be confused, you use the Claude Chrome extension to use the WeWeb editor or the app in production?

Either way, would it be possible for you to share the claude skill?

Best,

4

yeah you use it for both editor and prod, in my case i use it in the editor for it to fact check its work before i review or publish especially when you’re working in published apps. and the skills are saved in a doc folder

5

In my case i use Claude to open the editor to check the work it has done, and fixing errors. Sometimes it recommends publishing to staging, so it can test there as well. Its really helpfull.

For the skill its just part of a larger skill where i have described how my app is set up, what auth I use, what backend i use and how I communicate. Also UX details belong here. The skill also comunicates different things that saves on context length in each session.

You build this skill yourself using claude. Just tell it you want to create a skill with weweb best practices, your details and how you want it to follow up each change by opening your app in the web browser to validate its changes.