Jun 23, 2025 · 37m · startup-ideas
I built next gen AI Agents with MCP + Claude (STEAL my workflow)
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
Greg Isenberg and Riley Brown demonstrate how to build powerful AI agents using Anthropic's Claude and Model Context Protocol (MCP) within Docker. Through live demos, they show how to connect Claude to Notion databases and visual automation tools like Glyph to streamline complex business workflows.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Greg holds 12.4% of the talking time here. How this is scored →
Greg as informed peer 2.1 Guest teaching 4.5 Guest disagreement 0.6 Greg pushing back 0.7
speaking balance: gold is Greg, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 34:46 Failure of imagination remark
Riley strongly rebukes critics who dismiss MCP setups due to current friction, arguing that refusing to see past early jank is a failure of your own imagination.
Hardest push from Greg ▶ 34:23 Calling out Docker MCP jankinessGreg directly challenges the practicality of the workflow, highlighting how mind-bogglingly janky and buggy restarting Docker and quitting apps remains.
Biggest teaching moment ▶ 4:47 Perplexity and Cursor tool architecture lessonRiley educates Greg on how multi-billion dollar AI products succeed not through proprietary base models, but by wrapping LLMs with targeted external tool loops.
Greg holds their own ▶ 32:24 Notion as a headless database insightGreg offers his own expert analysis on using Notion strictly as a headless API database to bypass its sluggish front-end UI.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Greg as informed peer | Guest teaching | Guest disagreement | Greg pushing back | Why |
|---|---|---|---|---|---|---|
| Defining AI Agents and Tools Beyond Buzzwords | 1 | 5 | 2 | 0 | Riley immediately dismisses the term MCP as a distraction and buzzword, educating the host and audience on viewing them simply as agents using tools in a loop. | |
| Why Context and Tools Matter: Perplexity to Cursor | 2 | 6 | 1 | 2 | Greg asks why ordinary listeners should care, prompting Riley to deliver a breakdown of how Perplexity and Cursor built massive valuations simply by attaching internet search and codebase retrieval tools to standard models. | |
| Sponsor: Finding High-Growth Opportunities with IdeaBrowser | 3 | 3 | 0 | 0 | Greg introduces IdeaBrowser and notes how API key management is itself a startup opportunity, while Riley demonstrates Docker's MCP toolkit catalog. | |
| Live Demo: Integrating Notion MCP with Claude | 1 | 6 | 0 | 0 | Riley demos connecting Claude to Notion via Docker to search short-form hooks and generate new ideas, while Greg admits he feels overwhelmed by the setup. | |
| Comparing MCP Tool Capabilities with Claude Artifacts | 2 | 5 | 1 | 1 | Greg probes whether Claude Artifacts are obsolete or complementary to MCP tools, leading Riley to explain the difference between UI rendering canvases and persistent database tools. | |
| Glyph App Integration: Building Visual AI Workflows | 1 | 6 | 0 | 0 | Riley demonstrates how Glyph allows visual workflows to be packaged as mini AI employees callable directly within Claude chat prompts. | |
| Remixing Workflows and The Value of Good Examples | 4 | 3 | 0 | 1 | Greg asserts his view that most users will not build automation workflows from scratch but will remix existing templates, which Riley confirms using his thumbnail PDF example ingestion. | |
| Claude Projects and Notion Token Configuration | 1 | 6 | 0 | 0 | Riley walks step-by-step through configuring Notion integration tokens, restricting workspace permissions, and setting Claude Project system prompts. | |
| Why Notion Works Best for AI Agent Automation | 5 | 4 | 2 | 3 | Greg provides informed critique on Notion's UI sluggishness versus its utility as a headless agent backend, and pushes on the extreme jankiness of Docker MCP. Riley argues that ignoring janky early tech is a failure of imagination. | |
| Vibe Code App Pitch and Episode Conclusion | 1 | 1 | 0 | 0 | Greg wraps up the interview by having Riley pitch his startup Vibe Code App and praising Riley's pedagogical clarity. |
Statements from this episode (17)
Riley Brown: MCP is a buzzword distracting from agents with tools
“I think, personally, that MCPs is kind of a buzzword. It's distracting you from the sauce, and it's distracting you from the core ideas that you should be focusing on, right?”
Riley Brown: Claude Sonnet and Opus are agents out of the box
“Claude Sonnet IV and Claude Opus IV are literally agents out of the box, and so what we're talking about is giving tools to agents.”
Riley Brown: No company has made connecting tools to AI easy yet
“No company, even to this day, has made it really easy to add external tools.”
Riley Brown says Perplexity built its business on a single search tool
“Perplexity 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 othe…”
Riley Brown says MCP and n8n are the best AI tool frameworks
“I think MCP and N-A-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.”
Brown: Unified AI tool key management is a massive startup opportunity
“That is one of the biggest startup opportunities right now. I think people understand this. I think people will just realize that this is a much tougher problem than people think.”
Brown: Curating example databases for AI context is a major startup opportunity
“And so, in terms of, like, a, in terms of business ideas, some of the best business ideas might literally come from collecting good examples. I mean, look at this example, right? Like, AI never gave me good hooks until I created this little hooks database, so …”
Riley Brown reveals xAI is building a Canvas-like workspace feature
“Gemini has a Canvas feature. I know XAI is working on a Canvas feature, so they're all kind of adding this.”
Riley Brown: LLM canvas and artifact features are overrated
“I personally think At this point, they're a little overrated. Like, I hardly ever use them.”
Riley Brown says Glyph.app is the easiest no-key workflow builder
“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.”
Greg Isenberg predicts the future of automation is remixing existing workflows
“I don't believe that most people are actually going to go create NAN workflows, make workflows, Zapier workflows. Like, 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.”
Brown: Giving AI agents tools without instructions degrades performance
“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 going to get confused, and it might …”
Riley Brown: Highly Flexible Note Apps Undermine Team Collaboration
“I'm not a huge fan of like note taking apps where you have a lot of creative freedom because anytime you try and collaborate things become unclear. Notion, there's just too many options and bells and whistles in my opinion”
Riley Brown: Every Action in Notion Is Programmable Through MCP
“Google Docs makes it impossible, but Notion, literally every single action you could take in Notion is now programmable through this MCP. So an AI agent can just do those things directly if it has access to the Notion MCP server with all of those different too…”
Riley Brown: Tech Leverage Comes From Mastering Janky Early Tools
“And I think 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, right?”