How to Build & Sell AI Tools Without Coding Using Pickaxe

Full video transcript

Nathaniel, thank you so much for tuning in here with us today.

Hey everybody, it’s really great to be here. I’m super excited.

Nathaniel, for context for everyone, is the creator and founder of a company called Pickaxe, which helps you create no-code AI tools. They’ve combined a lot of the different things that have slowed a lot of people down from building and monetizing their own knowledge and expertise, and created a simple, easy-to-use platform you can use to either give people free access to your tools or gate it with paid access. There have been a lot of big updates over the past few weeks that Pickaxe has come out with that Nathaniel is going to share with us. So welcome aboard, Nathaniel. The stage is yours.

Awesome. Yeah, well, I guess I’ll just give the basic explanation for folks. I think Darby said it really well. If you’re out there building what you might call GPTs or agents or chatbots across a wide variety of contexts, people run into the same kind of problems again and again. They build this thing and it works well, but they can’t put it where they need to put it. They can’t really control access to it. They can’t really see what people are doing with these tools. So we built Pickaxe to address that, particularly for coaches and consultants, people working with educational institutions, Fortune 500 companies, small businesses, and individuals getting coaching or even therapeutic assistance. What you want is the ability to build these bots and put them into a box where you can brand it and control access. That’s really what Pickaxe is all about. So I might give a 10,000-foot overview of how to use Pickaxe and the different angles, and then get into some of the cool new stuff we’ve released recently — because our goal is to be the platform that allows folks trying to monetize GPTs, in one way or another, to do that effectively.

Love it. Yeah, let’s do that.

Yeah, so you would start with our GPT builder, which for those of you who’ve built one in an OpenAI system, looks quite similar. You’ve got your prompt-engineering area where you’re describing what you want your tool to do, your knowledge base where you’re uploading documentation, and action connections, which include MCPs and other things you might want to connect. For those who haven’t heard, MCP is a cool new framework that lets your AI tools connect with lots of other tools without setting up complicated API integrations — instead you click a button and it can do whatever you need across, for example, Notion or Google Drive: add, remove, edit files. So that’s where you’d set up your AI tools, and we have tens of thousands of folks who’ve built tools in this builder. But then Pickaxe gets interesting where you can deploy your tools as embeds — if you already have a course platform or an LMS, you can embed these tools right in there and “AI your courses” within five minutes, customizing the embeds to look the way you need for your business. One recently rolled out feature I’m excited about — I’ve never seen it anywhere else — is the ability to ask people to log into your embeds. Once they do, even on some other platform, you can see who’s doing what, limit usage based on the user, and take away usage based on the user, all of which you probably need if you’re taking AI seriously in a consulting practice. The other big thing is what we call “studios” — essentially a little website where we handle everything for you. With the click of a button, in about 10 minutes, you can go from building tools (in this case, a little murder mystery manor with several characters helping folks solve a mystery) to having authentication built in, so folks can sign up and sign in and you can grab their emails and understand your customers better, plus payments, so they can buy upgrades or get access to more tools, and limit them by credits to restrict usage — all built into the platform. So if you’re trying to monetize access to agents, this is a one-click solution from building to deploying into a branded, fully customizable environment, even at a custom domain, so it becomes part of your site. That’s the high-level overview of Pickaxe. I wonder if I can answer questions, or I might go into a couple of cool use cases we’ve seen, including what we’re starting to see in education.

I’d love to talk more about the practical ways you’ve seen other Pickaxes use MCP, because that was a newer addition I was trying to figure out how to get APIs to do before, and I wasn’t able to make it click. In my own experience with partners, we’ve figured out how to connect Pickaxe with a make.com automation and Zapier to export new studio signups to a CRM — in that case, Go High Level. There’s a lot you can do with this as a hub for tying different tools together, which I think is still kind of discombobulated with some other platforms. John also mentions, “I use Gum Loop to do automations but cannot gate it. Could I embed a flow like that or an n8n flow into Pickaxe?”

Yeah, that’s a great question, and we see a lot of that. With Pickaxe you can build a lot of complex automations, but we don’t subscribe to the idea of supporting really complex automation flows ourselves, because a lot of our customers use a tool like Gum Loop or n8n or Zapier or Make, and it’s very easy to think of this bot as the face of your automation — even something as simple as clicking a button that says “start” kicks off the automation. We have integrations for Gum Loop and n8n, and it’s easy to set up that connection; once you’ve got it, you can gate the automation and essentially sell access to it through Pickaxe. It can look like you’re just creating your own little chatbot with different bots here, but in reality each could connect to a different, quite complex automation under the hood, and you’re selling access to it through the action section. That’s something we see folks doing to great success. And to double down on Darby’s question — we’ve been advising people to use Smithery, though there are other MCP aggregators out there. One thing I’ve done myself is use the Notion MCP that Smithery has. It’s very easy to get set up, just a couple of clicks.

Could you show us how to do a setup? Are you able to?

Yeah, totally. Awesome.

Yeah, let’s do it. Okay, so here we’re in Pickaxe. We want to connect this MCP server, so we’ll do “add MCP server.” We’re going to get the…

Smithery, for clarification, is kind of like a bridge between some of these other platforms for authentication and helping make these MCP connections.

Exactly. So that just brought me over here and I had it add a new one, and I’ve got this key here, which I won’t show, but it took me about two clicks to get. I stick that into Smithery and click connect, and it gives me this stuff, which doesn’t matter much, but I’ve got it. Now I paste that in here — it’s writing the JSON code from all the API documentation you need from Notion to connect through Smithery, that you copy and paste to create a new MCP server in your Pickaxe.

Exactly. And now this has access because I added the API key to Smithery, which is hosting this MCP server — there are other tools that can host an MCP server, but Smithery has a lot of good tools already connected. Now I’ve got this Notion MCP connected. So if I say, “What can you tell me about Pickaxe pricing from Notion?” — I think I need one more step, which is clicking the connect button. I’ve added it but now I need to connect it. Let’s try one more time; it might take a minute here.

So you’re asking it this question because in theory the Notion page you set up has all the context about your pricing on it, so it’s pulling reference from that as a knowledge source.

Exactly. Let me just name this tool and publish it — “Notion searcher” — and let’s upgrade this model to something a bit nicer.

Some of the use cases here — I’ve seen people asking how they can update the knowledge base their GPT is referencing. All you need to do is update the Notion document, and Pickaxe should be communicating with the most current version, versus having to download all the information and re-upload it into the knowledge base. This is another way to grab that knowledge, and it’s more up to date.

Exactly. If you see this shimmering effect, that means it’s going and grabbing the information properly. It might take a minute, but we find people using all kinds of creative solutions for a knowledge base beyond what they used before, because they have access to these MCPs. People are putting information in Airtable, Notion, Google Drive — a sheet or a series of files — and MCPs can search through it. In this case we gave it the most basic possible one-minute setup, so it has to go find this stuff, but we could tell it later, “Go to the pricing page and find this stuff” behind the scenes so we don’t show that process to the user. This is a very quick example of one way to set up MCP, but inside the instructions where you put prompts, you can be incredibly specific about how you want these tools to work and provide keywords so specific phrases trigger different actions.

Bruce is asking, “As MCPs become more universally available, will tools like Smithery and other MCP hub aggregators replace the need for tools like Zapier and Make? Do you have an opinion on that, Nathaniel, or do you see them working together still?”

Well, it’s a really good question. I think Make has taken the world by storm a little bit, and I guess n8n as well — it has grabbed this information, took a little while, but it has grabbed the pricing information from our Notion, and it’s also telling us some maybe unneeded information about the way our Notion is structured, so in our prompt we might give it more instruction on exactly what we want it to talk about and what we don’t. We’ve seen Make and n8n adopt the MCP wave quickly, with Zapier tagging along from behind. I’m honestly not sure whether aggregators like Smithery will team up with them — maybe there’ll be an acquisition. I think it’s also a lot easier to host an MCP server than to build a traditional integration, because a lot of these are open source, so you might end up in a situation where instead of using Zapier and Make and other tools, you can just do a lot of this yourself, or there’s a platform that makes it easy to host these things alongside what you’re doing. With Pickaxe, we host some of our own MCP connections, like Google, Semantic Scholar, OpenAI Web Search, Sequential Thinking, and Google Sheets, and we’re adding more every day. So I’m excited, because it means everything is going to be so interconnected — your tools can talk to all the other pieces of data they need to get what they need done.

And you asked, do MCPs work more slowly than direct APIs because of the extra layer involved? They do work more slowly than direct APIs for sure, but it can be a benefit too, because what it just did there — this process under the hood — is actually making a series of API requests and processing what it got back, realizing maybe it wasn’t done yet, and going to grab another bit of information. That would have been a lot of work for somebody to set up themselves. Most of our customers can set up manual API connections — we have an area where you can code and add Python to set up a series of API connections if you want — but that would be too much for a lot of our users, and this just makes it super easy. It is slower, but if I went into this prompt area and said, “Always go directly to the pricing page in my Notion, don’t search further,” and added a couple of basic components to the prompt, I could reduce the amount of time it takes and make it more like a regular API call, and also customize what these loading states say, which can be nice since people expect to wait a little if it’s going to do something complicated. That’s an example of the power of — well, really we’re trying to talk Pickaxe here, and I recommend giving it a try because we try to make this easy for folks. But in general, if you find yourself scratching your head over complicated API integrations, checking out MCPs and playing around with them — I think that’s the technology of the next five years. So find a place, either Pickaxe or elsewhere, to experiment with them.

Yeah, that’s great. We’ve been talking about MCP since the last summit — it’s come onto our radar more, and learning how to use it and the technology in this space is still relatively young. But at the end of the day, I see MCP sticking around as a way for these tools and platforms to communicate better with each other. One of the cool things with an interface like Pickaxe — and I’d encourage you guys to check it out if you haven’t tried it yet, if you see value in gating your own knowledge or monetizing these types of workflows, Pickaxe will help make a lot of that process easier. There may or may not be other tools that do something similar, but specifically with Pickaxe, I appreciate that you can also set user memory and other settings that give you more ability to understand what people are going through so you can improve the bot. A lot of people I see using it are doing it on the coaching side, where they have a specific series of prompts that guide somebody from the start of a conversation all the way to the end, sometimes working with people actively throughout the process. Can you talk more about what’s going on behind the scenes in a studio, and then we can get into pricing and how some other stuff works?

Yeah, I’m happy to go through this, because in our goal to be the best one-click, easy way to monetize these GPTs, once you get them together into a studio like this with accounts and monetization, there’s a lot of other cool stuff you can pull out. And, Darby, as you mentioned, user memory is one of them. Not only can you have full view over what conversations are happening — we can go in here and look at, for example (I’m using the murder mystery as an example), what people are talking about, who they are, other conversations this person has had, a summary of who the AI thinks this user is. In a coaching solution, this summary might look at all the conversations and determine, “I think this person is running this type of business,” and you can export all of this to the CRM of your choice or view it in Pickaxe. You can also control this user memory ability — much like if you’ve used ChatGPT, you know it collects memories about you, but the thing about ChatGPT is it just collects whatever it feels like, and there’s really no control over what it remembers, unless you tell it, “Remember this.” In Pickaxe, if you have a coaching solution with a bunch of GPTs, you can decide what things it remembers about your users, and you can decide exactly how you want to show that to the user, or whether to show it to the user at all. For example, here I’ve got the user’s name set to collect and use, so if the user ever says their name, it will collect that and every new chat it will say, “Hey John, how’s it going?” That’s a very basic one. I’ve also got the suspected culprit — who they think did it — set to collect, use, and display, which is a choice that allows an end user of my tool to see their own memories and potentially edit or delete them. That’s important if you want your tool to remember a lot about your customers but also let them see and manage what information has been collected. One option a lot of folks use, that I really like, is just collecting — say I’m doing a therapy solution and I want to make an assessment of the user’s mental state every time they talk to the bot, I can set this to collect and configure it to be about mental state, and it will collect these without using them in conversation with the user, but I can see them as the owner of the tool. That’s helpful because, say the user’s mental state is bad one day, we probably don’t want the bot to remember that and constantly remind them every time it talks to them, but we want a record — when I click on John, I see that he’s having a hard time. I’m using a therapy example, but we’re seeing people embed these tools on their website for lead gen, and to the user it just looks like a regular AI conversation with a directed GPT answering a question, but behind the scenes it’s collecting rich information about the customer’s ICP, their likely buying trajectory if we’re trying to sell them a course or consulting service, and their key problems. So we can set up bots to record this information behind the scenes and use it to better understand our customers. That’s one reason the user memory area is really just the tip of the iceberg of what we want to do — the goal is to make it so that behind the scenes of these suites of GPTs you can build, you can build rich tapestries of customer information and understand your customers better than ever before. Maybe I can move over to John Gallagher’s question and answer some questions about monetization and access limiting.

Okay. I’ll share a link — it’ll be below this video wherever you’re watching it on YouTube. We connected before the summit and got 10% off your first three months, or 10% off your whole year if you go annual, set up for Gen AI University members who want to test Pickaxe out. If it doesn’t work for you, you can cancel your account. We’ll continue doing more trainings around Pickaxe and demonstrations, and are happy to answer questions between summits. One example that came up in our mastermind months before MCP was announced with Pickaxe was frustration with not being able to have a GPT automatically update whenever you update a Google document in your knowledge base — you’d constantly have to re-download it into the knowledge base, which was time consuming, and there wasn’t an easy workaround. So similarly to the Notion demo, is that another application — could you build an MCP to a Google document so a Pickaxe interacting with it always gets the latest version whenever the document is updated in the cloud?

Absolutely. That’s actually why we have this Google Sheets MCP noted right here. A very common use case is a document, or often a Google Sheet, that’s being updated and they want the Pickaxe to pull from it over time — and sometimes a separate document the bot writes to, which they’ll review periodically to see what questions people are asking and go find answers to stick back into their other documents. We’re trying to make those flows even easier, and some of them you can do right in Pickaxe. Our goal is to be a flexible crucible for all these different workflows, providing an easy-to-use, no-code visual layer where you handle access limiting and user management — the “last mile” stuff that’s difficult to do with some other platforms. One thing I want to highlight is an unreleased feature I have access to, which we’re excited to release soon, probably early next week. I’ve talked about embeds and studios; you can also deploy these tools as APIs. But newly, we’re going to allow you to deploy these tools as SMS.

So you can have a phone number for your tool and people text in and chat back and forth with the GPT. Just like everything else on Pickaxe, it works the same way — you’ve got a usage limit, you can encourage people to upgrade or pay through Stripe, and come right back to that SMS conversation on their phone. So these GPTs are coming out of the websites they’ve been living on and into SMS, and the next thing we’re going to do is WhatsApp. I’ve literally been texting back and forth with my Pickaxes all week. I think this is really going to expand the opportunities to monetize, because now these tools can reach people without them even having to open a browser.

That’s super exciting to me, and I didn’t know this was coming — this is the first time I’m hearing about it. I’ve tested and experimented with setting up WhatsApp agents before, and there’s a lot of configuration involved in building your own WhatsApp chatbot; this is going to make that much easier. To clarify one workflow: if I could connect SMS with a number, I could text my Pickaxe bot — let’s call it Moxy, my AI operating assistant — and if Moxy has an MCP to my Slack channel and other resources, could I text it and say, “Can you check recent DMs or app mentions to me in my Slack channel?” and have it identify messages and relay them back and forth over SMS?

Absolutely — I love that question because it pulls together everything we’ve talked about today. If you’ve got the MCP set up through Slack that can write, read, and search, you’ll be able to talk to your bot and ask exactly those questions, and if it’s set up through SMS, all the MCP power we’ve been packing into the web version of Pickaxe is going to be available there too. I haven’t been using it for our Slack yet, but I do have it set up to correspond with our community, and I’ve been

using a tool that connects me to what’s going on there — I can get updates and even post responses all from this tool. That’s been a lot of fun. It’s a little slow if I’m honest, but it’s been very fun to do.

That’s really cool, and I think it’s inspiring to see the direction agentic interactions and interfaces are going. It’s great that it’s on the web, but everybody has a phone, everybody texts, WhatsApp is one of the most used apps in the world, especially outside the US. Being able to engage directly with your own AI assistant, or one you create for a client, removes friction, because it means not having to jump between all these different platforms — which is why we want to connect these different knowledge bases and communication platforms. I’d encourage everyone watching to consider what tools you’re already using and get organized about what you’ll need — an MCP for Notion, an MCP for Slack — so when it’s time to set one of these up you’ve already done that work, especially with SMS and WhatsApp coming. If your workflow isn’t mapped out, now would be a good time to spend an hour putting it together, the same way you’d think through what access a new human assistant would need before asking them to do all this stuff. Really cool stuff, Nathaniel.

We find it’s good to start small. Oftentimes people jump onto the platform, set up something really simple, and then call us the next day saying, “This works so well, now I want to connect it to this and do this other thing with it.” You can’t know how easy it will be until you give it a try, and once you try it for one self-contained thing, that’s when the ideas really start to spark. We had a conversation yesterday with a major university about signing a deal to allow all of their professors to use Pickaxe in the classroom. It started with one professor using it for one specific type of assignment, then using it for other things and sharing it with other professors, and now the school is reaching out wanting a bigger deal — it really started from one person with one small idea. I’m hoping that happens more often, because if you’re the one having that idea and proliferating the technology, you’re making everyone you work with more powerful.

Yeah, and so much of it starts with just one simple use case — what’s one workflow or outcome you’re already doing that you could save time on or do a little faster? That’s different for everyone, but once you find those use cases, there are probably other people, clients, or community members who’ll benefit from learning what that looks like too, which is why we host community events and the Scale With AI Summit — there’s always new advancements in this technology. I love connecting with people like you, Nathaniel, who are building the platforms other people use, because you have a different perspective on how people can connect the dots and see practical value from these tools. Thanks for coming on and sharing with us. For those of you here, go check out Pickaxe — there’s a link wherever this video is posted for 10% off your first three months or 10% off your first year. Thanks so much, Nathaniel. Thank you.