The argument in one line.
AI agents become exponentially more useful when given access to external tools through Claude Projects and Docker, which lets you automate complex workflows like content creation by combining Notion databases, AI image generation, and web search into a single orchestrated system.
Read if. Skip if.
READ IF YOU ARE…
- A solo creator or small team running content operations who wants to automate repetitive tasks like thumbnail generation, database management, or content organization without hiring engineers.
- Someone building a product or service who needs to understand how AI agents with tools actually work in practice, beyond the MCP buzzword, through a concrete working example.
- A non-technical founder or operator who wants to see how Claude Projects can orchestrate multiple tools and databases to handle workflow automation at scale.
SKIP IF…
- You're already comfortable building AI agents in code or have a technical team — this is specifically designed for non-technical setup and understanding, not advanced implementation.
- You need a production-ready, fully polished system today — the video explicitly acknowledges the current setup is janky and still evolving.
- You're working with fiction, creative writing, or domains where AI agents with database-connected tools aren't directly applicable to your workflow.
The full version, fast.
MCP is a distracting buzzword for what actually matters: giving AI models tools they can use in a loop, which is the working definition of an agent. The mechanism shown is Claude Sonnet or Opus connected through Docker's MCP toolkit to external tools � Notion for context-aware content workflows and Glif for remixable mini-workflows like thumbnail generation � wrapped inside a Claude Project whose system prompt tells the agent which databases to read, where to comment, and which glyphs to invoke. The practical conclusion is to stop building n8n workflows from scratch and instead remix existing ones, curate a library of strong examples to feed the agent, scope each integration's access narrowly, and accept the current jank because the leverage compounds as tooling matures.
Members feature
Chat with this breakdown.
Modern Creator members can chat with any breakdown — ask for the hook, quote a framework, find the exact transcript moment. Unlocks at T2: refer 3 friends + add your own API key.
Create a free account →Who's talking.
00:00hostGreg Isenberg
00:58guestRiley Brown
Where the time goes.
00:00 – 01:08
01 · Intro & teaser
Greg solos to camera, frames the MCP promise and difficulty of setup, introduces Riley Brown.
01:08 – 04:36
02 · MCP = agents with tools
Riley strips the buzzword. Defines AI agents as models using tools in a loop. Introduces n8n. Uses Anthropic definition slide.
04:36 – 08:42
03 · Why tools matter
Perplexity built $1B on one tool. Cursor is Claude with code-search tools. Boring Marketer applies same pattern to content creation.
08:42 – 10:34
04 · IdeaBrowser ad read
Sponsored mid-roll for ideabrowser.com.
10:34 – 14:00
05 · Docker MCP Toolkit
116-tool catalog. Riley shows Notion and Glif integrations enabled. More tools without matching instructions = confusion.
14:00 – 22:56
06 · Notion + Claude live demo
Riley demos querying his Agent Mind Notion workspace, pulling short-form hook structures, searching web for Dia browser, generating three content options, writing entries back to Content Database — all in one Claude chat.
22:56 – 28:07
07 · Claude artifacts vs MCP + Glif intro
Greg asks about artifacts. Riley explains artifacts as a render layer, distinct from MCP tools. Transition to Glif.
28:07 – 32:45
08 · Glif thumbnail workflow demo
Riley shows Glif.app — visual workflow builder, no API keys needed, public + remixable. Runs Thumbnail Ideator from Claude via MCP. Uses PDF of example thumbnails as style reference. Outputs five options on a canvas.
32:45 – 36:00
09 · Becoming the AI orchestrator
Riley shows VibeCode CEO Claude Project — system prompt referencing Notion for rules, [[double bracket]] triggers for Glif workflows. Human role becomes quarterback.
36:00 – 37:15
10 · Janky now, inevitable soon + VibeCode pitch
Honest take on current friction. Riley argues leverage comes from building with janky tools before they are smooth. VibeCode one-liner close.
Lines worth screenshotting.
- MCP is a buzzword — the actual concept underneath it is simply agents with tools, which is all you need to understand.
- AI agents are models using tools in a loop; they decide what tools to use and for how long, which is why they are not the same as automations.
- Perplexity built a billion-dollar company by adding exactly one tool to an AI model — the ability to search the internet before responding.
- Cursor is so powerful because it wrapped a familiar model with the right tools; the model itself is the same Claude everyone else uses.
- Giving Claude access to your Notion database as a tool means it can generate content with real context instead of hallucinated context.
- Most non-technical people can build powerful AI agents today using N8N because it makes the agent decision loop visual.
- The gap between people consuming AI and people building with AI is the same gap that has always existed between audiences and creators.
Build your Agent Mind now, while the tools are janky.
JoeFlow + creator stack
The window to build a content workflow that compounds is open right now — precisely because most people are waiting for it to be easier.
- Create an Agent Instructions document written for the AI, not for humans. This is your context moat.
- Pick two MCP tools max to start: one for knowledge base (Notion, Obsidian) and one for output layer (Glif, or a script).
- Set up a Claude Project as your orchestrator. Give it your rules, triggers, and voice.
- Build a hooks/examples database and give AI access to it. The database is the edge.
- Steal Riley's double-bracket trigger pattern for firing specific workflows mid-chat.
- JoeFlow framing: Sessions + Chef = the same orchestrator pattern, native to the desktop. Pitch: you already know this works — here it is without Docker.
Terms worth knowing.
- MCP (Model Context Protocol)
- An open standard for connecting external tools and data sources to large language models. It lets a chat assistant call outside services like Notion or a web scraper instead of being limited to its training data.
- AI agent
- A language model that uses tools in a loop, deciding on its own which tools to call and for how long until a task is finished. Different from a fixed automation because the steps are chosen at runtime, not pre-wired.
- n8n
- A visual workflow automation platform similar to Zapier where users chain triggers, conditional logic, and AI agent nodes together. Popular for building no-code agent workflows.
- Zapier
- A no-code automation service that connects apps through pre-built triggers and actions, used as the reference point for how integration marketplaces work.
- Claude Sonnet 4 / Claude Opus 4
- Two tiers of Anthropic's Claude model family used as the underlying agent. Sonnet is the faster mid-tier and Opus is the most capable, both designed to handle tool use natively.
- Cursor
- An AI-powered code editor built on top of VS Code that wraps Claude with tools for searching the codebase and the web. Used here as an example of an agent made powerful by its toolset.
- Docker MCP Toolkit
- A catalog inside Docker Desktop that exposes pre-packaged MCP servers, letting users enable external tools for an AI chat app without setting up each integration by hand.
- MCP server
- The background process that exposes a given tool's capabilities to the AI model over MCP. It must be running locally for the chat app to see and call its tools.
- Composio
- A platform that aggregates third-party integrations and exposes them to AI agents, positioned as an alternative way to add tools beyond Docker's catalog.
- Firecrawl
- A web scraping and crawling service that can be wired into an agent as a search-the-internet tool. Requires its own API keys to configure.
- Glif
- A no-code workflow builder where users assemble chains of AI models and image steps into reusable mini-apps. Each workflow can be called as a tool from inside Claude via MCP.
- Notion integration token
- A secret key generated inside Notion's developer settings that grants an external app or AI agent permission to read and write specific pages in a workspace.
- Claude Artifacts
- A Claude feature that renders generated content like code, landing pages, or diagrams in a side panel of the chat. Comparable to OpenAI's Canvas and Gemini's Canvas.
- Mermaid diagram
- A lightweight text-based language for describing flowcharts, sequence diagrams, and other visuals that get rendered into images. Useful for letting an AI produce diagrams as code.
- Claude Project
- A workspace inside Claude where users save custom system instructions, files, and rules that apply to every chat in that project. Used here to give an agent a persistent role and toolset.
- System prompt
- The hidden set of instructions given to a language model before the user's message that shapes its behavior, persona, and tool-use rules across a conversation.
- ComfyUI
- A node-based interface for building generative image pipelines, referenced as the kind of complex visual workflow builder Glif was initially mistaken for.
- Vibe coding
- Building software by describing what you want in natural language to an AI rather than writing code by hand. The term has expanded into adjacent areas like vibe marketing.
- SOP (Standard Operating Procedure)
- A documented set of repeatable steps for performing a task, stored here in Notion so an AI agent can read and follow the same process a human team would.
- BYOK (Bring Your Own Key)
- A model where the user supplies their own API keys for each underlying service the tool calls, instead of paying a single bundled subscription. Common pain point when wiring up many MCP integrations.
Things they pointed at.
Lines you could clip.
03:00
“AI agents are models using tools in a loop.”
05:30
“Perplexity built a billion dollar company adding one single tool.”
34:56
“Most of the leverage from these tools comes from doing things when they're janky and bad, understanding that it's not going to be janky and bad.”
35:38
“Anthropic, I know you're watching. Plug it into the platform. It'd be great.”
13:05
“I want to hire someone full time and all they do is just pull really good examples.”
Where the conversation goes.
01:08 – 08:40denseMCP concepts and definitions
08:40 – 14:00steadyDocker MCP Toolkit walkthrough
14:00 – 22:56denseNotion MCP integration live demo
16:57 – 20:10sparseClaude Artifacts digression
28:07 – 32:45denseGlif workflow builder live demo
32:45 – 36:00denseClaude Projects as orchestrator
36:00 – 37:15steadyCurrent state and future of MCP tooling
Word for word.
metaphoranalogystory
00:00I don't know about you, but I heard that you can put MCPs within LLMs like Claude, and you can get a 100 x more out of them. The problem is it's kinda difficult to set up. So I brought on Riley Brown who's gonna go through a tutorial with how to set it up in a simple way so that you can actually transform your business, save hours of time, uh, and outcompete 99.9% of the people who are using AI today.
00:32You're not going to want to miss this one.
00:43Riley Brown back on the Startup Ideas podcast. What are we going to learn today? So today,
00:50we're going to be using MCPs, but I don't wanna frame it like we're using MCPs. I think, personally, that MCPs is kind of a buzzword.
01:00It's distracting you from the sauce, and it's distracting you from the core ideas that you should be focusing on. Right?
01:08What does MCP mean? Doesn't matter. For the sake of this video, doesn't matter.
01:13What we're talking about is agents with tools. And if you look at the trends right now with, uh, if you look up MCP on Google Trends Greg, I know you love, uh, Google Trends. Um, if you look up the trends for n eight n, if you look up the trends for the next biggest AI agent builders, people are dying to build AI agents.
01:37And by the way, Claude Sonnet four and Claude Opus four are literally agents out of the box. And so what we're talking about is giving tools to agents.
01:49That's all I would really wanna talk about today because it is so significant. It is so important to basically any workflow.
01:58And the one that's going the most viral right now is n eight n. So what n eight n is is it's a kind of like Zapier where you can create these workflow automations, except you can create this AI agent node.
02:11And AI agents are really hard to build. Right? I am working at a startup where we're building an AI agent, doing it in a code base is impossible for nontechnical people to, like, wrap their heads around, like, what it's actually doing.
02:26And I think what N8N did is they made this interface for AI agents so that nontechnical people can understand them, where instead of it being, like, a workflow, um, I think I, uh, included it.
02:40Yes. Right here. It's not if one thing happens, then the next thing happens, then the next thing happens.
02:46Maybe you have some conditional logic. Right? You're you're, um, you have a lead or you have an email, and depending on what type it is, maybe it sends it down one path, and then it sends you result a or sends result b.
02:59Instead, something happens, and then it sends it to an agent, and it has access to tools.
03:06So the agent will decide what tools to use and for how long. Right? An AI agent, if you've used Cursor, it might spend ten minutes on a a prompt, or it might spend thirty seconds on a prompt depending on how much, uh, resources it needs to allocate to that specific task to get a specific result.
03:27And so the best definition for AI agent that I've ever heard that kind of removes all of the extra noise because so many people think automations are agents. So many people think the the the definition for AI agent is very elusive right now.
03:43And so AI agents are models using tools in a loop. Right? It can you it can spend as much time as it wants in this step.
03:52It can use tools over and over again until it's done, and then it will send out this final result. Right? That's how I see it.
03:59And so today, we're gonna be talking about MCP and giving Quad, which is an agent, access to tools.
04:08Right? So you get to choose which agent model you wanna use, SONNET four or Opus four, and then we get to actually add tools, which if you have used Claude, it's this little button right here. And we're gonna talk about how to actually give it access to tools, which has actually been the biggest barrier so far.
04:25No company even to this day has made it really easy to add external tools. I'm gonna show you one company that has made it somewhat easy, but it it has a lot of room to get easier. Is this making sense so far, Greg?
04:38It makes sense, but I have, like, a dumb question to ask, which is, like, why why do why should the person listening to this
04:46podcast care?
04:47Absolutely. Okay. So as you know let's actually go over here.
04:56Tools matter a lot. Right? If you think of the original ChatGPT and you asked it to create content about some event that happened two days ago, ChatGPT would have just hallucinated some random script not in your voice, and it would also not have context surrounding whatever it is that you're talking about.
05:19Perplexity built a billion dollar company, right, in the beginning, adding one single tool, basically, which was search the Internet before responding. They built a billion dollar company doing that, and this gave its responses incredible context where no other AI model had.
05:40Now, uh, all of these tools have a search tool, and so now it's no longer their sauce. So they're searching for other things. They're trying to create other products for their sauce.
05:49The same thing with Cursor. Why does it matter that you have an agent with tools? Right?
05:54Cursor is a really great example of a really powerful AI agent in an and by the way, it's a very familiar AI model. Right? It's Claude.
06:03Most people use Claude. And it is a familiar interface with, um, Versus Code.
06:10But what they did is they created tools so that it could search the code base, which was one of their tools. Now it can search the Internet. So you can say search the Internet, find documentation on whatever AI feature you wanna create, and it will implement it correctly.
06:23It's amazing. It's a miracle. And that's because of tools.
06:26Right? And so then why is boring marketer talking about MCPs and NA to end workflows?
06:33Right? So the same way that Cursor surrounded a smart AI models with tools it needs to create apps the user wants, boring marketer is surrounding smart AIs with relevant tools so he can create content, copy, um, and different things like that, like anything that's relevant to marketing.
06:52Right? So he's creating his own code base, if you will, uh, of, like, marketing topics and giving AI agents access to YouTube summarizers, AI ad copy creators, maybe the doc like, documents to his voice, um, allowing it to scrape other people's websites.
07:11And I've come around to the term vibe marketing and just because I have to, I just have to accept that it it's just gonna win.
07:19And I think MCP and n eight n, the reason why he's talking about this is I think it is the two best ways right now to kind of give AI agents tools. Okay.
07:31At least for this specific workflow because we're gonna be using Docker.
07:36Sam Altman, the cofounder of OpenAI, just said that it is the era of the idea guy, and he is not wrong. I think that right now is an incredible time to be building a startup.
07:47And if you listen to this podcast, chances are you think so too. Now I think that you can look at trends to basically figure out what are the startup ideas you should be building.
07:58So that's exactly why I built ideabrowser.com. Every single day, you're gonna get a free startup idea in your inbox, and it's all backed by high quality data trends.
08:11How we do it? People always ask. We use AI agents to go and search what are people looking for and what are they screaming for in terms of products that you should be building, and then we hand it on a, you know, silver platter for you to go check out.
08:26Um, we do have a few paid plans that, you know, take it to the next level, give you more ideas, give you more AI agents, and more almost like a chat GBT for ideas with it. But you can start for free ideabrowser.com.
08:39And if you're listening to this, I highly recommend it. So Docker is traditionally it's like a pretty technical
08:46platform. Um, but I I just ended up seeing that they created this MCP toolkit.
08:55And since I already had Docker on my computer, it made it really easy to just add tools. So if you if you go to Claude right here, you'll see if you're just starting out on Claude and maybe you have it downloaded for the first time, you're not gonna see anything down here.
09:11You're just gonna see nothing. In order to give it access to tools, you need to, like, you need to go to Docker or any other tool. I think Composeo does something similar.
09:22I I'm not specifically it doesn't matter which platform you use. Docker's just the one that I use.
09:29And so if you go to MCP Toolkit, you can go to catalog, and they have a 116 tools that you have access to. And they're adding a ton more apparently.
09:38And so this is kind of like, you know, Zapier. Zapier has, like, so many integrations with all these different tools, and I think that's where this is headed, where basically all of these external tools that you might need are just gonna be in a giant catalog.
09:53And so, for example, like, we could add FireCrawl, which searches the Internet. For this this one, you're gonna need a ton of configuration keys, so I may not use this one right now, and that's where this is still annoying.
10:05I can't wait for the client that comes in and handles the keys for something like FireCrawl, then we just pay one company instead of having to set up accounts on all of them. Which is a startup idea in itself. Like, why doesn't right?
10:19A thousand percent. That is one of the biggest startup opportunities right now.
10:24I think people understand that. So I think people would just realize that this is a much tougher problem than people think, um, connecting all of these together, um, for many reasons, which we don't need to get into.
10:34But I wanna talk about my two favorite tools that I've used so far. The first one is Notion, and the second one is Glif.
10:43And this Glif tool is literally like creating AI employees, and you can basically create an AI agent that has access to little AI employees.
10:54And I'll explain that more in a second. But let's start off with the Notion tool. Right?
10:58So whenever you wanna enable a tool on on on Docker, once you enable it, on Notion, there's, like, this little integration token you need.
11:07Right? You'd basically need to give it your account's key so it knows which account that it's using, and you can very easily find that.
11:15So let's go ahead and open up Notion. And my this I have a Notion, uh, team space that's literally called agent mind.
11:25It's made for AI agents. And so the home page of this and by the way, this is where I keep a lot of my content stuff.
11:33It's the the top of the page is, uh, here are your instructions if you are reading this, AI agent. And so it's made for AI agents to be able to navigate it because with MCP, it can do all of the tasks that I do on Notion manually.
11:48It can do, um, with MCP. And, again, MCP is just giving an agent tools for the sake of this video.
11:57And so what we can do is we can let's go half screen here. And when you use Docker, you make sure that it is, like, still on.
12:08Like, don't exit out of it. You get access to the tools, but it needs to be running. That's the MCP server running just by having Docker open.
12:17And so when I go here, I will see all of my Docker tools that I've enabled. And so I have all of my glyph tools and all of the Notion tools, which is a total of 31 tools. So these are all of them that it can do.
12:29And so now we can say something like so in Notion, I have a content I have a how to create content SOPs.
12:36Right? I have this page right here. And so this goes over, like, short form hooks, long form hooks.
12:42And so I can ask it. I can say, um, look at, um, at my short form hooks in this database and write what's an a tech news thing that came out recently?
13:07Dude, I'm so overwhelmed. Oh, Dia browser. Dia browser.
13:10Oh, yeah. Short
13:13form hooks in this database. Write short form content about the new Dia browser, search the Internet first and and find info.
13:33Right? In this case oh, wait. I need to tell it to use Notion because we haven't set up our custom prompt.
13:38So use Notion MCP, and I'll show you how to avoid having to do this every time in just a second.
13:47Use Notion MCP.
13:51Find the database below and, um, write three options.
14:00So right now, we're just giving it instructions, and there's no system prompt for this. So right now, it's using a tool. Right?
14:08More and more of these chat apps, like ChatGBT, Claude, have tools just built into them. Right?
14:15Claude, three weeks ago, didn't have search or maybe a month ago didn't have search. Now it has search. Here, it's actually doing API post search, so it's actually finding see.
14:24Perfect. I can see there's a short form hooks page. Let me grab the contents from that page and understand the hook structure.
14:30Okay. I see there are tables with hooks. Let me grab the content of these tables to see the hook structures.
14:35Great. Now let me search for information about the Dia browser. So it searched, um, all of my all of the available Notion database, and then it found the how to create content database.
14:49Then it found this short form hooks, and now it's giving us options.
14:56So here are the two options. Right? And we can see this.
15:01Today, I realized that browsing the web is about to change forever. Right? If we go into short form hooks, we can see that one of those hooks is today.
15:10I realized that. And and so in terms of, like, a biz in terms of business ideas, some of the best business ideas might literally come from collecting good examples.
15:23I mean, look at this example. Right? Like, AI never gave me good hooks until I created this little hooks database.
15:31So whenever I'm scrolling, like, if I find a hook that I like, I'll just write it in this database really quickly. And that in and of itself could be a huge company if you can, like, provide the right context.
15:43Maybe you have, like, some custom system instructions at the top. Like, if you're writing I don't know. You you have to find the sauce.
15:50And so it's all about finding the right output, giving really good examples to get that output. There's a lot of really good business ideas in that space. Right?
15:57The next one was this feels illegal to know. And so I'm sure that's on here. This feels illegal to know.
16:04Right? And so it's pulling that. And so now what we can do is we could just say, like, can you please add these as separate, um, entries in my content database?
16:21And now, um, it should, in theory, add them to the content database because it not only can read things from it, it can also, um, paste things.
16:33It can type things. It can add comments, which is my favorite feature actually is being able to add comments on what I'm doing so it doesn't change what I've written, but it'll, like, provide feedback and give me options.
16:46Is this all making sense?
16:48It makes sense. One thing I was thinking about is, you know, how how should we think of MCP versus Claude artifacts?
16:57How should we think of MCP versus Claude artifacts? So Claude are so it just added it here.
17:04So you can actually get Quad. You could make a custom project within Quad and give it instructions that, like, anytime it creates an artifact, you could get it to also upload it to whatever database you're working in.
17:19I've always found that that for me, like, one of the biggest things is whenever I create something with AI, it kinda gets lost. And I especially if I, like, edit it or change it, I like to store it somewhere.
17:30So that's one thing that's really useful for me is being able to quickly just tell AI to store it somewhere. Artifacts is obviously this separate feature.
17:40I I don't they're not like I don't know if they're complementary, but Claude basically just renders your chats on the side window. And you can share them.
17:50You can edit them. I think you oh, you can't edit them on Claude directly. You have to, like, highlight it and tell AI to do it.
17:57On OpenAI's kind of version of this is their Canvas feature. Gemini has a Canvas feature.
18:04I know XAI is working on a Canvas feature, so they're all kind of adding this. I personally think, at this point, they're a little overrated. Like, I hardly ever use them just because
18:16I don't know. Do you use the artifacts on quality? That's kinda that's what I was getting at really, which is, like, I'm trying to find a way to use artifacts just because it's it it looks like a core feature, but I'm kinda like,
18:30I don't know if I'm dumb for not using it. Like you know? So Claude Artifacts was one of the most impactful features for me because it allowed me to, like, render like, say, like, create front end component and or, like, create a landing page to mock up and build it, and it would render it right there.
18:49It was the first time because Claude Artifacts was the first one to do it. Now there's so many tools that do that, and they allow you to add AI features, and they allow you to add back ends that, like, it's just not what I really wanna be doing.
19:02I know that, um, in the episode that I did with Boring Marketer, he had a really cool way of doing it where he's using MCP, and he's bringing in information. Right?
19:11So the same way that I'm searching the Internet, he's he's deeper in the trenches in terms of, like, analytics, like, where he'd, like, scrape Reddit for specific things. And then he would take the information and render it in in the you know, like, he would say something like, please create an artifact of actually, we could say, create a mermaid diagram of all of the new features of Dia.
19:46Right? And so you can actually go out, find external information, and this will actually just, like, render this in here.
19:52And Mermaid is Mermaid is a very cool, like, little coding language that renders this code in a really cool way.
20:01You're gonna see it in a second. It's very cool. And it makes information more visual, which I love.
20:11I'm very visual. And, of course, this is, like, the wackest diagram ever. That's, like, horizontal.
20:16But but but if by the way, if you copy this and you go to a tool like if you go to if you ever use Mermaid, you can go to mermaid.live is a much better renderer.
20:35Right? This code right here is rendering to this.
20:39This is still gonna be pretty horizontal, but it makes it easier to, like, see it, in my opinion.
20:47And you can edit things directly. But it it just made, like, a really horizontal diagram, which makes it, like, really hard to read because it included everything. But you can see here that how do I get rid of this thing?
21:01But it's, like, d a first browser. You can see that it's, like, a personalization engine, etcetera, etcetera.
21:07So that's just one little side thing. But what I wanna get to, what I really wanna focus on is the idea of having little workflows that you can run like they're little AI employees. So if we go me go back to this.
21:24So let me go to glyph.app. This is a tool that I just found randomly that, like, I think is so is the easiest workflow builder out of all of them, and you don't need external API keys.
21:37Like, I actually couldn't believe that because I've been hearing about this tool, I thought it was just, like, kind of like like some comfy UI builder.
21:46It's not. So here, let me show you. So what I created the other day was this so there's this workflow.
21:55Right? And these are really easy to make. You can edit.
21:59So, like, you have, like, an input image. And here.
22:04Let me go like this. I'm just gonna there's Greg.
22:10So now we can drag Greg into there.
22:15Right? And we can just run this glyph.
22:19And so it's gonna go through this workflow. Right? And there's a step in here called fix ratio.
22:25I didn't create that. Because glyphs are default public, you can just constantly remix other people's, um, flows.
22:33And why am I showing you this? You're like, alright. Why is he using a different tool?
22:36Because on the shut up. See?
22:42And you can just, like, create these little workflows, and so you can make them this this transfers them to be vertically oriented. And yeah.
22:51And so you these little workflows are called, like, VibeCode hack. So after you have set up the Docker integration, right, after you've set up the Docker integration for Glif, then in Claude, what we can do is let's just go ahead and open up a new chat and be like, what Glif workflows have I created that I can use?
23:19And this uses the MCP, and it sees all of your glyphs, all of the glyphs that I created.
23:26That one was called vibe code hat for my company. There's another one called you'd get YouTube thumbnails. And so here are so we have this, like, thumbnail ideator real photo realistic.
23:38And so I'm making a video on, um, using Docker for MCP.
23:48Please use this, um, to make a thumbnail, And then you can include, like, it empowers the Claude agent for better agent, so it's 10 times smarter.
24:13Right? We can add some, like, buzzwords in there. And so we can just run this.
24:19And let me see if this works. Okay.
24:22So it's running that same glyph. So right now, it's doing the equivalent. Everything that we could do in here where we, like, go because we would have to click here.
24:31We have to say your glyphs. We'd have to find thumbnail ideator, and we can do that. I still do that all the time, but it's very nice to be able to just use them in the chat interface.
24:41And you can tell it to use if you have, like, four different workflows, you could say run all four of them with this one prompt, and it will just run them. And, like right and so what it's doing, this is a longer workflow where there's like it takes your idea.
24:59And by the way, you could say search the Internet, come up with a good prompt for it, and then give it to this workflow, and it will go through this long process. And this is my thumbnail ideator thing.
25:10So it basically generates five different thumbnails and puts them on a canvas with a title so I can get an idea of which one I wanna do, and then maybe I'll screenshot my favorite two and send them to my thumbnail guy. And that's kinda how I come up with ideas.
25:25And it's doing this process here, and it's also doing it on quads. Any questions, Greg?
25:33It's it's funny because Glyph, I've seen, but I always wrote it off. I don't know what it what is about it, but I do think like, going through this now, I mean, this is I don't believe that most people are actually gonna go create n eight n workflows, make workflows, Zapier workflows.
25:54Like, I don't think that they should. Like, I think the future of all of this is you're just remixing other people's workflows.
26:01Yeah. So, like, this to me makes a lot of sense. No.
26:03I 100% agree. Like, I'm trying so I've been, like, learning NaN, and I've actually gotten, like, quite good at it. But I agree with you.
26:11Like, it actually requires, like, some pseudotechnical knowledge that, like, I just don't believe. So look at this.
26:17Like, it went through this full workflow, and here we have five thumbnails that we can choose from. So, yeah, that's the output of Lyft. Like, what do you think?
26:26Like, these are pretty good thumbnails, right, for just, like, quick ideation thumbnails, and it also puts a title there as well?
26:34I mean, do mean? These are these are dope. Like, I I I spend a lot of time I spent some time coming up with thumbnail ideas, and, like, these are, you know I think I'm good at it, but, like, these are as good if not better.
26:47And you wanna know what's funny? You wanna know what's funny? Is you are
26:51part so remember I talked about examples earlier. Right?
26:56We talked about really good examples. That's my secret now. I collect examples.
27:00Like, I wanna hire someone full time, and all they do is just pull they're just finding really good examples. So part of the input of this Claude step right? So you just type your idea, and there's I don't this I can make a whole video on this.
27:14So I'm not gonna go too deep into the details. But one of the things is that it is looking at this PDF. And so, like, what is this PDF?
27:22Right? It's downloading this PDF right here, and we can see what it's doing is it's looking at these thumbnails that I've saved.
27:32And since Claude you it doesn't just, like, convert this into text. It can actually ingest a PDF, see all of them, decide which ones it wants to make, and then it uses it as an example here.
27:43And so, obviously, we have some Callaway, Cleo Abram. I think she has thumbnails.
27:47Shout out Greg Eisenberg. You you know, like you and so it's basically this and and so, yeah, and so I just give it some really good examples.
27:57You can make this as long as you want, and you can make this as sophisticated as you want. However, I always recommend starting with simplicity if you're just starting out.
28:06Like, make it very simple. Maybe only use Greg's thumbnails, and also make it use his face and and just rip it. No.
28:13I'm just kidding. But yeah.
28:15And then it goes through all of these steps. And so if you get really good at these workflow builders, now you can start thinking of yourself as, an orchestrator of an AI agent that also has access to this.
28:27Because the the what I'll close on is this. Right? You're like, okay.
28:33You're probably thinking, okay. This seems cool. You can you can set it up on Notion so it can read things.
28:37It can create things. You also have this workflow builder. So if you're kind of, like, working on an idea, you can pass your ideas to this workflow builder.
28:45What you need to do like like, sounds kind of annoying to have to type that in every time. You can create projects.
28:51And so I've created one called VibeCode CEO. Right? And what it does is, like, you're the CEO of VibeCode.
28:58You answer questions in order to create content. It's like the the CEO of the content team.
29:04And so it says you have access to Notion. This is the most important tool you have access to. Use it.
29:11When asked to make content, please comment on the actual blocks or when asked to comment on wait.
29:23Found a little mistake in there. And then I also have these, like, little fun little things where if I ever put double brackets around something and I were to say, uh, resolve comments, that's just my thing where it's just gonna go through the current Notion page, and it's gonna look for single brackets and double brackets.
29:41If it sees double brackets, I want a thumbnail. So I want it to use that thumbnail glyph, and it will paste the link into the comments.
29:50And so it'll literally just, like, doink. It'll paste it in there. It won't crowd the page.
29:55The AI will comment on it. And, um, and you'll even see that whatever you name your Notion integration, which I actually will show you that real quick for those of you who actually wanna just, like, practically try this, which I recommend.
30:08It's the best way to learn. If you go on Notion and just hit settings and you click on connections and you click on this develop or manage integrations, this is the key that you're gonna need to plug into Docker.
30:22Oh, yeah. I set up Dia. Dang it.
30:25Yeah. I did not want that to be my default browser.
30:33Right. Okay. So I have one called MCP Anj, who's my cofounder.
30:39I have one called n eight n. And, yeah, and so you can basically just create a new integration, and maybe you want this to be your content in integration.
30:49And then your associated workspace is gonna be the company. But I'll show you if certain times you may you may want to give certain MCP integrations access to only specific files within your code base.
31:05Right? You don't wanna give it you only wanna give it the marketing stuff. And so in access, right, you can hit select pages, team spaces, and that's why I've only given it agent mind.
31:15Right? There are other team spaces in there with, like, important sensitive information that we don't wanna give up, or it just wouldn't be relevant to the AI agent. It won't help it get any better.
31:25So part of your job as kind of the quarterback of the AI agent with access to tools is you need to decide what tools it has or hasn't. The more tools you give it access to without corresponding instructions, the more it's just gonna get confused, and it might actually make your life harder.
31:41And, yeah, and so you can give access to this. And so this token right here, um, I would hit show and copy, but I obviously don't want you to have access to my Notion.
31:51That'd be weird. You just go into Notion. You go to configuration, paste that key right there.
31:58Then every time you add a configuration, what you need to do is you need to go to clients. You need to disconnect from Claude.
32:06Quit Claude anytime you make changes, and you just basically need to restart the server. So you can just restart it back up, and then any changes you made to your MCP tools will actually be updated.
32:20And Does that all make sense?
32:23It does. Two two quick thoughts. One is we use Notion.
32:27I've been using it for years. I actually hate Notion. I find it sluggy sluggish and just, like, yeah, sluggish and slow.
32:35But all this MCP stuff is making me appreciate Notion again. Like, as long as I'm not spending time in the interface that much, like, I don't really care. That's the first thing.
32:47Yeah. No. No.
32:48Do you wanna go I guess I'll to respond to that, first of all, I'm not a huge fan of Notion. I'm not a huge fan of, like, note taking apps where you have a lot of creative freedom, uh, because, uh, anytime you try and collaborate, uh, things become unclear.
33:01Notion, there's just too many options and bells and whistles, in my opinion. But it is by far and I've tested all of these tools.
33:09Like, Google Docs is kind of a nightmare to do these similar work. You have to set up all these special integrations. You need many, um, see here.
33:18It says you could not attach to MCP Docker. Very interesting.
33:23But, um, uh, Google Docs makes it impossible. But Notion, literally every single action you could take in in Notion is now programmable through this MCP.
33:33So an AI agent can just do those things directly if it has access to the Notion MCP server with all of those different tools. You have to enable all of them, but you can't. Like, every single thing you could do, it can do in Notion.
33:44So, yeah, you now have this little AI agent that can do it. And if you set very specific, um, Claude project rules or, um, you put you can actually put the rules directly in Notion.
33:56So you can just say read the rules in Notion, then do your thing. So, like, make like, you just type into the system prompt in Claude projects to always follow the instructions on x page.
34:06So it'll look at that first, and then it will follow those instructions. Everything will stay organized. And you you're right.
34:12You basically never need to touch it. You just kind of write whatever you want in Claude, do research in Claude, and everything gets organized perfectly into your Notion, which is pretty cool.
34:23The second point I was gonna make is I think everyone should play with this. Set this up.
34:28Do it yourself. Like, stop listening to us. Like, go and go and get your hands dirty.
34:32At the same time, I think it's mind boggling how janky this experience still is today, like with the Docker and like, oh, you know, it gets disconnected. Oh, quit the app and then reopen it.
34:44It's it's bonkers.
34:46It's really bad. And I think I think most of the leverage most of the leverage from these tools comes from doing things when they're , understanding that it's not going to be janky and bad.
35:02Right? I was interested in AI coding or now it's called vibe coding when it was really bad because I knew it would get really good. It got really good, and and you can, like, grow with the progress of technology.
35:14If you can't see the potential of this technology of, like, making your normal chat interface have access to tools, like, that is a failure of your own imagination.
35:26At least that's my opinion. Right? Because it is inherently useful to ground it in context that can search the web better.
35:33It can scrape the Internet. It can look through your Notion. It can do all these things.
35:36Is this process janky? Yes. But will it be solved by the end of the year?
35:40I think for sure. Like, I think Claude might build it in directly into their platform. They should.
35:45Anthropic, I know you're watching. Plug it into the platform. It'd be great.
35:50And make it easier. Like, make it as easy as possible on people to add these tools. And if, like, if Claude were to do this first, like, they could take a lot of the market share from, like, the prosumer crowd.
36:02I I truly believe that. Like, it's that important.
36:05We can we can end the pod here. Thanks for being so generous with your time. Riley, I know by the time this is out, VibeCode app will be live.
36:16People could can you can you give a quick one liner on it? We're building the, uh, cursor,
36:21uh, on your phone that builds mobile apps. So you open a mobile app, you type your idea, it builds a mobile app, and it's awesome.
36:30I've been working on it ten hours a day to twelve hours a day for three and a half months straight. We're building a team in San Francisco.
36:38It's gonna be awesome. Check it out if you want. Um, Yeah.
36:42We're doing a launch video. It should go out tomorrow if all goes well. We filmed it yesterday, so it's a very tight time loop.
36:47But, yeah, thanks for having me on. I mean,
36:52Riley, not to make you blush, but I have I have a lot of respect for Riley and how simple he's able to explain things and how generous he is with all all he's doing and all the content he's putting out there. So highly recommend you follow him.
37:06I'll include social links where you can go ahead and do that. And, um, hope to see you soon, Riley. Appreciate it.
37:13Thanks, Greg.
The Hook
The bait, then the rug-pull.
The hook is a permission structure: Greg admits he does not know how this works yet, which gives every non-technical viewer cover to keep watching. Riley Brown enters as the explainer, armed with a live Claude session and a Notion workspace he has literally written for AI agents to navigate. By minute eight, the buzzword is dead and the demo has begun.
