MCP onboarding is the most exciting thing happening in tech right now.
Moe Khalil
SubscribeSign in
MCP onboarding is the most exciting thing happening in tech right now.<br>Excuse the clickbait title. Obviously, I know there are more exciting things happening right now.
Moe Khalil<br>Jul 17, 2026
Share
But over the past few weeks, my co-founder and I have been putting in a lot of hours into figuring out how to make our product’s agent experience as seamless as possible, and oh my goodness the deeper I got into it the more exciting the problem became.<br>All of a sudden, your user is not interfacing directly with your product. There’s a messenger in the middle, capturing your user’s intention and resolving it onto a surface your product offers. Suddenly, everything we know about good UX is thrown out the window. The touch points, the edge cases, potential causes of frustration -- they occupy an entirely new problem space altogether.<br>Thanks for reading! Subscribe for free to receive new posts and support my work.
Subscribe
So as I’m finishing up this work, and having recently launched our MCP to the public, I thought I’d share what I’ve learned along the way.<br>1. Onboarding belongs in the agent, not on your website
The point of an onboarding process is to teach your user how to use your product. The best examples of this have always made using the product part of the onboarding experience. For some reason, this hasn’t caught on as much as it should with MCP. MCP education is still largely in the form of API-style docs, which puts the onus on your user to spend time on your website parsing through them -- which I promise you they don’t want to.<br>The way I approached this was by including an onboarding tool within our MCP, which gives the agent a step-by-step process that walks your user through their first time using your product. By the time this process is done, your user should understand exactly what your product offers, how they can use it, and should have gotten value out of it already.<br>2. Part of the onboarding process should be setting up your user’s workspace (with their permission)
Everyone’s machine looks different, and for a lot of products, everyone’s motivation behind using your MCP is different. Given how unpredictable agents are, and the fact that every new session starts with a new chain of thought, it’s so important that your user’s workspace is set up for them to get the most value out of your product.<br>Luckily, modern agents are built first-and-foremost to write code and operate a machine -- all they need is guidance on how. That guidance needs to be part of your onboarding experience.<br>Our onboarding process tells the agent to ask the user for permission to do a meta-analysis of their workspace, fully locally without sending anything to us, then gives suggestions on how to change their setup to get the most of our product. If the user says yes, the agent executes.<br>3. Agent-native support can be better than founder support
At this point, a lot of us have company brains -- and very, very well-commented code. An agent with access to all that probably knows just as much -- probably more, but I know some people won’t like hearing that -- about your product as you do.<br>If that’s not the case, just open your agent, brain dump everything about your product into it via your preferred speech-to-text, and then ask it to read through your code. It’ll be able to come up with something that’s significantly better than nothing.<br>Then, make that accessible to your user via the MCP. A help endpoint, with FAQ, product info, and anything else people might ask you about. Then make that a living document. Your users will thank you.<br>4. Uninstall is part of trust
I personally believe that the tipping point from prospective user to power user is at the moment of first trust. Earlier, I wrote about the importance of asking the user for permission. That’s a big part of that. But the even bigger part, especially when you’re changing your user’s local workspace, is making those changes very simply reversible.<br>Oftentimes, the difference between a product that works and a virus is how easy it is to get rid of it. So make your MCP as easily uninstallable as possible. For us, that’s an endpoint similar to our onboarding endpoint, which walks the agent through getting rid of our MCP and all associated rules, and uninstalling the MCP completely.<br>Having said all this, it’s important to remember that agent-first interfaces are still in their nascent stage, and all this may come to be disproven. If that’s the case, I’ll be the first to admit I was wrong and update our product to fit whatever works better. I invite people who disagree to share their thoughts; I’d love to hear them.
Moe<br>CTO at Webhound<br>Thanks for reading! Subscribe for free to receive new posts and support my work.
Subscribe
Share
Discussion about this post<br>CommentsRestacks
Ready for more?
Subscribe
© 2026 Moe Khalil · Privacy ∙ Terms ∙ Collection...