May 11
Loading video player...
š n8n Desk (GitHub): https://go.n8n.io/n8ndesk-github š n8n Desk (Website): https://go.n8n.io/n8ndesk-website What if you could use Claude Code to build n8n workflows and never worry about it running destructive code on your machine? In this episode, Marcel joins the show to demo two things: n8n's brand new official MCP server (currently in beta), which lets AI agents build workflows for the first time, not just run them. And then his own open source project, n8n Desk: a free desktop app that looks and feels like Claude Desktop, but routes all code execution through n8n instead of your local machine. That means encrypted credentials, a full audit trail, support for any AI model (including local ones), and role-based access for enterprise teams. We also get into vibe flowing, context engineering, and why n8n is actually easier to debug than AI-generated code. Timestamps 0:00 - Claude Code destructive actions story 1:40 - Guest intro & today's topics 1:45 - New ways to build n8n workflows 4:05 - Connecting n8n MCP to Claude Code 5:01 - Old MCP vs. new MCP 5:25 - New MCP builds workflows 5:33 - Any agent can create workflows 6:47 - What is "vibe flowing"? 7:41 - Why n8n is easier to debug 11:32 - AI fatigue & reclaiming your time 11:37 - Setting up the n8n MCP connector 12:50 - Building an IT support buddy agent 13:22 - Vibe coding vs. context engineering 15:14 - Agent creates workflow inside n8n 17:31 - Why the first attempt failed 17:35 - MCP ships built-in reference skills 19:22 - Community MCP & combining skill stores 20:43 - Custom skill stores & plugins 23:13 - Skills as .md files explained 24:21 - Latest & greatest, expect improvements 24:29 - Marcel's side project teaser 24:42 - The problem with Claude Desktop 28:24 - Introducing n8n Desk 28:30 - Run any AI model locally 30:34 - Cowork tab: local files via n8n 37:20 - Chat Hub: talk to n8n agents 38:03 - Kelly calendar agent live demo 42:20 - Free & open source release plans