Stop Making These 5 Webflow MCP Mistakes
Watch on YouTubeAbout this video
Are you getting the most out of the Webflow MCP?
In this video, I break down 5 common mistakes people make when working with the Webflow MCP, and how avoiding them can help you get faster, more accurate results from AI.
You'll learn why it's important to:
- Connect your entire Webflow project, not just a single page
- Review every AI-generated output before publishing
- Start fresh instead of endlessly prompting the same conversation
- Give your AI session enough context to work effectively
- Set realistic expectations for what AI can (and can't) do
Whether you're building websites, managing client projects, or exploring AI-powered development workflows, these tips will help you work smarter with the Webflow MCP.
Chapters
- 00:00 Introduction
- 00:21 Mistake #1: Limiting AI to One Project
- 01:09 Mistake #2: Trusting Every Output
- 02:39 Mistake #3: Prompting in Circles
- 03:29 Mistake #4: Not Giving Enough Context
- 05:06 Mistake #5: Expecting Too Much from AI
- 06:29 Final Thoughts
Full transcript
1,584 words
All right, everybody. Today we're going to be talking about the top five mistakes people make when working with the Webflow MCP. My name's Emmett and I'm a Webflow professional partner. I've been working in Webflow for 5 plus years and I've been working with the MCP a lot recently. I'm going to give you some tips for what I see people do wrong and how I think we can fix some of those things.
The first tip is connecting to a single site only. The Webflow MCP can connect to your entire workspace and that's much better than connecting to a single site because if you connect to the whole workspace, you can see every single site within that workspace and you don't have to reconnect, disconnect, reconnect, disconnect every time you want to change or work on something different. Especially for a freelancer like myself, I'm constantly moving around and working with different tools.
So, make sure that you're connecting to the workspace instead of just a single site. If you are working across multiple workspaces, what I do is put a scoped MCP folder inside my project and then when I switch to a different folder, that new MCP is available. So, it's basically creating copies of the connection so that my tools can always be up to date and I don't have to log in dozens of times a day.
That's mistake number one. Mistake number two is not reviewing the output of your tools. So, what I like to do is do a dry run. When I'm working with the MCP, what I'll say is do a test first with one object, one item in the database and let me review before you go and make 200 changes. Get all the context that you might need. Make sure that you understand this collection. Move one item. Make one change.
Let me review first and then once I approve, then it can go forward and do 200 changes or whatever it may be. That's especially useful if you're doing a bulk update because last thing you want to do is have to undo and restore a backup and deal with your client upset that you've changed hundreds of items or that your AI has all of a sudden gone wild on your site. It's very important that you don't just let your tools run wild.
I like to do a dry run and to test if it's working properly. It's even more helpful to do a dry run in a spreadsheet if you don't necessarily trust the direct connection. You can export your content in a spreadsheet, do the test locally on your machine, and then upload that imported file. Another thing that I like to do is to start a brand new session and then ask it to compare item one and item two or the previous before and after because the tools are really good at a specific task.
So, you can say, "Does this look the same as this? Does A look the same as B?" That's a lot more useful than saying, "Did you do this correctly?" Of course, it's going to say it did it correctly, but it doesn't actually know. What it does know is item one matches item two or the before matches the after. So, make sure you review the output of whatever tools you're using. Mistake number three is prompting in a circle instead of just starting over.
AI tools really can only go forward. They don't really have the same concept of history or going back in time that you and I do. Even an undo key is really not a possibility for an AI. If you have your content backed up, then you of course you can go back, but the AI's memory can only go forward. Even if you just say, "Go back to what we just did," it doesn't quite realize what that is. I always like to start over instead of trying to go around in a circle.
You really have to think you're always moving forward. Would you rather try to go around in a circle to where you are now or would you like to just clear it and start over? It's almost always faster to just clear the deck, start over, abandon your current session, and then use your backup or whatever tool you have to go back in time and then just start over. That's another benefit to just working in a draft mode or a branch page.
Just delete it and start over. Mistake number four is not giving your tools enough context. It's important when you're working with the tool to prep the session, start the session by telling your AI to look at your whole site, look at the classes, look at the variables, and understand what's available for it to use. Especially if you've worked with an agency or a freelancer, chances are that you have a style guide with variables and classes that are available, and your AI doesn't know that by default, even if it has connections and access, it really doesn't have a brain to understand that it should do that first.
So, you have to tell it to do that. Tell it to look at the site and to understand what's available, and even tell it to go to the docs for something like Client-First or Lumos, whatever naming convention you're using. Tell it to go there and find the classes that it should be using. So, that's important. The next step is to have your AI tools ask you questions until it's clear on the context. So, this is a trick I've been using for a while.
At the end of my prompt, I say, "Ask me questions if this is not clear or if it'll help you do your job better." Something like that. Because you may not realize that you've forgotten to say something important, or there's just something easy in your brain that you could be a little more explicit and make it a little more clear. Just tell your tool to ask you questions, and that can often solve that problem.
Then your tool will have enough context to do what it should do. So, make sure that the scope of what you're asking your tool to do is clear, and a great way to do that is to have it ask you questions. And lastly in this points, one thing that you can do is to be explicit about what not to do, as well. Especially if you're not wanting it to default to some sort of AI slop or something that is very obvious, like a slider.
And you want an accordion and not a slider. So, just be explicit and say, "Do not do this. Do not use blue. Do not use rounded corners." Be explicit. Mistake number five is expecting too much from the AI tools. You need to be more direct with the work that you're doing with your tools and not expect it to know or to assume what it should do. And that's a really hard problem because it basically requires you to think like a programmer and to think, "What technically am I asking?
Is it a two-column layout? Is it a four-column layout?" And those are technical language and technical things that you need to know. But it's very hard for the AI tools by themselves to understand what you need. And one of the most important things that you can do here is to be specific and to break it down into small specific chunks for the AI. So, you can say, "I need this one section, and it needs to be" or even show it an example.
It needs to look like this with these two columns, with a button here, and be as specific as possible. And that can really help it to be more effective. Things like nice, good, better, all those vague words are pretty much meaningless to any AI tools because those are subjective. And all it knows is something like, "Make it blue. Increase the contrast between the background. Make the font smaller." Those are all very specific requests that the AI would be a lot more capable to do.
So, that is really an important piece is to not expect too much and instead be specific and break it down into really small chunks. And that really can feel a lot like programming. So, I understand that that can be hard, especially if you're not a technical person. But, all those tips are going to help you avoid the mistakes. And all those tips are going to increase your effectiveness and to make you create better content with Webflow.
So, as a recap, the first mistake is connecting to a single site instead of to your whole project, not reviewing the content that you output with AI. That can be a big problem, especially when you're working with clients. Number three is prompting in a circle instead of just restarting and getting a blank slate. Number four is not giving your tools enough context to do their job properly. And number five is simply just expecting too much from your tools.
Make sure you're specific. Make sure that you're clear. And all those tips are going to help you be better. Thanks.