Welcome back. This is a very exciting lesson. We are going to set up Claude Desktop, although this will work with Codex, Cursor, whatever. We're going to set up Claude Desktop so that you can connect it to Bento and have your AI assistant from your tool, Claude Desktop, actually build and edit your emails in Bento for you.

So this is one of the coolest features of Bento. Really excited to walk you through it. Maybe a little scary, a little techy if you're not used to some techy stuff, maybe a little scary. And in that case, talk to Tanuki within the normal Bento interface, just like you've learned, and it can help walk you through this.

Or talk to your own AI assistant for generalized commands, like we'll talk about in a minute. I'll walk you through them. So, first thing that we'll be doing is pulling up the page in Bento where your API credentials are. I'm not going to actually pull mine up because I don't want you to see my API credentials, but I'll at least show you where it is.

So from within your Bento account, you will click the settings cog at the top right, and then API keys, and then you'll click Site keys here. I'm not going to click mine because it shows some of mine. And within that Site keys tab is where you'll see the stuff corresponding to this screenshot. If you don't have one yet, you'll generate a new one, and we will use it in a moment.

So once that's done, we will run one little terminal command, which is optional but helpful, and this is the part that I said might be scary if you're not super techy. This package here, so basically if you run this, there's this library called Skills. It's like a code library that allows people to share different agent skills. And if you don't know what an agent skill is, basically agent skills are just these instructional guides written specifically for AI agents of how to do certain kinds of things.

So Jesse made one about Bento so that if you install this skill, your agent will now know how to work with the Bento MCP, the Bento API, et cetera. Really useful when you want to connect to the Bento MCP. So when you run this command, what it does is, it installs the skills via the NPX Skills package. So you can talk to your normal old Claude or ChatGPT or whatever to understand how NPX Skills works.

I don't remember if I had to install NPM first to do it, or Node first to do it. Probably I did. It's been a while since I had to do the core scaffolding, so I really don't remember what prereqs will be in place. But as a first step, just try running this command and see what happens.

Okay, so instead of just letting you figure that out, I just had a little convo with ChatGPT about this, and indeed, you do need to have Node.js installed first. So [clears throat] if you don't have it installed yet, you can go to nodejs.org and download it. If you're not sure if you have it installed, I put a few little commands you can run to check. So you need Node, you need Node Package Manager, and you need NPX.

I don't know what the X stands for. [chuckles] And you can run these three version commands in your terminal to see if they're installed. So see how I ran these here, and the first one says V23, this one says 10.9.2. This one also says 10.9.2. That means they're all installed for me.

And if you're on Mac, to open the terminal, you can just do your search and type terminal. And then for Windows, I don't really remember. ChatGPT says it's called PowerShell. That makes sense.

So if you go to your search menu on Windows, type PowerShell, that'll open the terminal, and then you can just paste these commands to see the versions. So once you have Node installed and NPX, you can install the Skills package. And if you ever want to find out more about it, it's at skills.sh. This is the general repository for all skills, and I bet if we type Bento here, we would see it since he put it on here.

Yes. There you go. Oh, no, this is something else. So his might be private or something.

But in any case, this is what it uses. And NPX specifically allows things to be executed. So in this case, it's executing the adding of the skills. That doesn't really matter for what we're doing.

All that matters for now is that you run this and that you successfully get the configuration that I screenshotted here. So when you run that command, it will ask you a bunch of configuration questions. Given that I just ran it, I don't know if I'll be able to run it again. Let's see.

I was able to. Cool. So basically, you can use your arrow keys up and down to navigate this list. And what it has you do is choose which agents you want it to be available for by default.

And so see how it says space select, enter, confirm. So let's say I wanted it for Aiderdesk. See now it has this green thing. By default, it's going to install it for all of these.

So the only, in general, special one you might want is Claude Code. Let's see if there's a separate one for Claude Desktop. I don't think there is. Yeah, no, there's not, because this is just adding it to the .claude directory.

So for today's tutorial, we just want to make sure that that's checked off, so it should have a green dot next to it. Then you hit enter, and then you'll be given the option of a global or project level install. For what we're doing today, I think global is simpler. So I'd recommend you go the global route, and just hit enter.

And then this installation method, I would recommend the symlink thing. So what a symlink does, and I can show you, I've got it pulled up, it creates a little reference, basically. So see how here we're in .claude/skills? And this is on Mac, so it's different on Windows, but it does the same.

I think you could do symlinks on both operating systems. They might just render a little differently. But notice how this renders like a shortcut with this arrow. So what happens is when you go to claude/skills/bento, the actual file lives in agents/skills/bento.

So if I were to edit this file, I don't want to, but if I were to edit this file and change like this to be called bentos within this agents directory, when we were here in the claude directory going to claude/skills/bento, it would also be reflected here. So the reason this is a useful thing to do, the symlink, is that as AI tooling grows and you adopt it more, having all of your directories locked into Claude specifically or Codex specifically or whatever agent you use specifically is kind of a problem because now you don't have any portability between the different agents. Whereas if you always put them in a centralized generic directory and then just create symlinks to it, what it means is that suddenly it doesn't matter if you're using Claude or Codex or Cursor or whatever the heck you want for your agent harness because they're all routing into that centralized place. So this is why we do the symlink.

So we just hit Enter, and I don't know what's going to happen to me reinstalling something that's already installed, but we'll try it. So what you'll do first is you'll just confirm the summary that it looks good. The main thing you need to confirm is that you're indeed having Claude Code in the list since that's what we're using today. And this little squiggly line means your root user directory.

So in this case, see how it's like macintosh HD user Zach Swinehart. That's what this squiggly line means, is we're starting from the Zach Swinehart. And so everything after that squiggly line matches here. So agents, skills, bento.

See? Cool. And then we'll just hit yes, hit enter. And then it says it's done.

It gives you the directory, and this step is complete. So next thing to do is the MCP setup. So what we've done at this moment is we just installed a skill for an agent to use, but that agent can't yet actually connect to Bento until we set up the MCP server. So MCP stands for Model Context Protocol, which is a pretty useless term for explaining what it does.

If you haven't heard of it, it's basically an easier way for an AI to connect to some external tool than, say, making an API call. Which is, if you don't know what an API call is, it's a traditional way that one software tool can talk to another software tool. But you have to pass data in a very specific way, and MCP just makes it a little bit more AI agent friendly. You can look up MCP or talk to your agents about it to learn what it is.

But basically, TLDR, MCP makes it easy for your AI to connect to Bento, and we need to set that up specifically. So if you go to the Bento MCP docs, it'll walk you through how to do it for your specific tool. So you can choose your tool from the list. I'm doing it in Claude Desktop.

And definitely do read through all the steps on the page. But we've done the core ones already. We already grabbed those API keys, so now we just need to hook them up, and that's what I'll be walking you through in a moment. So if we go to Claude Desktop on here, that's what we're doing.

And fun fact, if you're on Mac, that I learned recently, if you want to easily pull up this path, instead of having to navigate manually to it, what you can do is open your Finder and hit Command-Shift-G and just paste it, and that'll navigate directly to this path. And so you'll open up this JSON file with your default text editor is fine. I'm not going to scroll down because I'll be showing you my secrets. But basically, in that file, so Jesse has a little code snippet here.

The file will likely be much longer if you're using Claude Desktop a lot than what's shown here, and that's fine. Just don't replace the entire file. We just want to put this one part here. So if you already have an MCP servers section in your file, you would only paste this part, basically everything within that, and you would need a comma after the previous one.

So if there's one in there already called like ABCD, and it's got its own little squiggly brackets, you'd add a comma after the ABCD one, and then hit enter and paste Bento remote below it. And then this part here is what you need to configure. So those environment variables are all available on this page I had you look at. You'll be able to find the site UUID, and then you'll also be able to find the API keys.

So you need all three of them, and you'll paste them into here, just like it shows on this page's instructions. And then once you have done that, save the file, close the file, and if you already have Claude Desktop or whatever agent harness you're configuring open, you should probably quit it and reopen in order for the tool to discover your newly added MCP tool. So with that done, the moment of truth now is checking if Bento actually exists in Claude for you to use. So if you open up Claude Desktop, you can see I've got this Bento chat already open.

What you need to do is go to the Code tab. Claude Code is a very important aspect of this because, in order for Bento to run MCP commands, use your file system, et cetera, you'll benefit from being in Claude Code. You could maybe do it in Claude Cowork, but I feel like they're in the process of deprecating Cowork anyway. So go to the Code tab, click New to make a new chat, and then basically, we want to make sure that the skill is available and that the MCP is available.

So if you just type Bento You should see it, and if you do, that means that this skill was installed correctly, so that's awesome. And then we want to check that the MCP was installed correctly, so you can click the little plus sign and go Plugins. And we want to make sure that we-- Oh, sorry, Connectors. My bad, not Plugins.

We want to make sure we have Bento Remote here. So if you see that, then it means everything worked well. If you don't, and you already did restart Claude, then something was messed up along the way, and that would be worth talking to your agent about, or better yet, talking to Tanuki about to try to troubleshoot, because most likely there was some skipped or misconfigured step on either the MCP page or the AI agent setup page. And then a couple other notes is that, number one, there's a troubleshooting section on the Bento MCP guide, and that you can always just reach out to Jesse on the Bento Discord if you get stuck.

He's very helpful. So once that's in place, if you want, you can run the setup prompt. It seems like it's a bit more strategic, this setup prompt, like what should we do inside of Bento kind of thing. But nonetheless, it's a good way to check that everything's working.

But another good way to check that everything's working, which is what we will do, is actually have it create some emails for us. So just grab this. And I'm not going to waste Opus credits on this. [chuckles] I'll just use the sonnet. So we paste it in here.

Send from, let me think of what my email is. This has to match the email on the account. I'll just go check what it is. So if we go into emails, maybe?

Somewhere we have the authors, yeah. Emails and then Authors. So it looks like it's coming from my zach@zachswinehart, so [bloop] we need to make sure it matches. If it doesn't match, it won't work.

Like if it tries to create a broadcast with an email address as the from address that doesn't correspond to an author in your Bento account, it will fail. So let's just see what it does, and I'll take a look at the next steps while it's working. So basically, agents have access to what are called tools via MCP, and a tool is essentially a way for the agent to perform some pre-made kind of code stuff but call it with normal human language. And so Create Broadcast is one of the tools that Jesse has made available in the Bento MCP.

Let's just do this. This is fine. And so what we want to keep an eye out for, so number one, see how it says using Bento Remote? That's great.

That's what we want to see, because that means it is using the MCP server, and you can expand it and get a bunch more info, and it says it created it, and it even gave me a link to it. But before I click that link, I'll just go browse for it. So if we go email/broadcasts, boom, look at that. Amazing.

Let's see the content. Cool. So that's great. Now we know that the agent was able to connect with MCP, and that's the biggest hurdle.

So that's awesome. Now we get to do the exciting part, which is building out a sequence with AI. I think this is where you really start to get a feel for the power of MCP connecting to Bento. So what we'll do is pull up the sequence we created earlier in the previous section of the course for the free plan upgrade sequence.

So if you go Emails and then Sequences, [clears throat] and then navigate into it, you can click the little Edit button and then copy out that name, and that'll allow the agent to look it up. But actually, probably you shouldn't have copied the name yet, because we first need to copy the prompt. So I have this little starter prompt you can use. So copy and paste that, and then copy the name, because we'll paste it into the prompt.

So replace this bit. Before I send it, let me make sure I didn't miss any steps. No, I did not miss any steps. So we'll send that.

And at the time of recording, if you try to just paste the URL, it won't be able to successfully look it up right now, because this URL slug does not match the actual ID. I did mention this to Jesse, and I think he's planning to change how this works a little bit so that it would be more flexible. But as of this exact moment, the most reliable way to look up a sequence is by name, not by URL. So just keep that in mind.

Don't blindly copy and paste that URL. So it says it did it. Let's go take a quick look. So you can see right now, by the way, before I refresh, the only thing in here was this first email I made that said, "Upgrade your plan, dawg." Now, if I refresh, we should have three proper looking emails.

Look at that. Wow. These were all created by the AI via MCP, and it just invented these based on the prompt, which is amazing. What's also cool is it can edit emails, too.

So it's not just for initial scaffolding, it can make tweaks. So let's have it do a targeted change. So we're going to have it change the second email. So let's take a moment and just look at what it currently says.

So no testimonial at all in mine right now. And now I'll copy and paste this prompt, and we should see it add or inject, rather, a new testimonial from Testy McTesterson. So we'll let it do its thing, and then we can refresh, and we will see that. So it says it did it.

Let's take a quick look. Here's what we see right now. In fact, I can maybe do a new tab so we can compare. Yeah, cool.

So as you can see, it's the same. It just added that testimonial. Amazing. So this, what you just did is pretty much a good proof of concept for everything you might do long term.

In real life, you might use this for quickly putting together sequences. You can have the agent do the drafts, like in markdown files for you to review. I personally like Obsidian for reviewing these kinds of files. This page you see in front of you right now is me reviewing a markdown file within Obsidian.

And you could get your sequence good and then have it create those emails for you. You could have it make iterations based on stats and performance, edit a bunch of emails at once, that kind of thing. Really, really useful. And just as a heads up, while we used MCP here, there are a bunch of other ways that you can work with Bento as well.

That's one of my favorite things about Bento is how many ways Jesse has created for you to interact with Bento from other tools. So you can work with the API directly. He also has CLI tools, command line interface, which will be useful if you're a developer, or for your agent to use. He also has SDKs and a bunch of third-party integrations.

So browse any of these that seem interesting to you. I use, for example, the WordPress plugin for my WordPress site, which would allow me to do things in Bento based on students inside of LearnDash, which is a plugin I use for managing my courses and stuff. So there are all these really cool things at our fingertips here that are easy to not know about if you don't go look. So maybe browse through the third-party integrations and just see what exists and see if any of these are tools that you use.

So here's my WordPress one. Maybe you use Stripe, Shopify, et cetera, and see if anything looks nifty for you. But for right now, for the scope of this, using it with AI, you don't need any of that, and we're all set. So at this point, if you did all the stuff I did here, you're good to move on, and I'll see you in the next lesson.