N8N just shipped a major update on August 11. For homelab admins who want to automate complex workflows using AI without sending data to cloud providers, the new AI Assistant and Agent Builder capabilities are worth paying attention to.
What’s New
The update brings three main additions that matter for self-hosters:
AI Assistant: Instead of building workflows by dragging nodes around, you can describe what you want in plain English. Tell N8N “Monitor my self-hosted RSS feeds, summarize unread articles using my local LLM, and post summaries to Telegram.” The system translates that into workflow nodes. This works because N8N’s parser now handles more natural language patterns.
Agent Builder: More ambitious than the Assistant, Agents let you compose multi-step AI reasoning. An Agent can look at a situation, decide what steps to take, and adapt as it goes. If you run Ollama or Open WebUI locally, you can wire Agents directly to your local LLM—no cloud API calls, no data leaving the network.
Model Context Protocol Support: N8N now implements the MCP handshake standard (finalized July 28, 2026). If your other tools support MCP—Open WebUI does—then Agents in N8N can discover what’s available on your network at runtime. You don’t pre-wire integrations; Agents see what’s there and use it.
The release also adds Azure Key Vault endpoint configuration (useful for on-premises deployments), improves PostgreSQL and MongoDB support, and makes the VM expression engine the default for all dynamic expressions.
Why It Matters
Zapier and Make are cloud-only. N8N is self-hosted. This update pushes the difference further.
Most N8N users today export their workflows to a server and run them. Agents change that. With Agent Builder, you can chain together dozens of tasks without manually wiring every connection. Throw in local LLM support (Ollama running on a homelab machine), and you can build sophisticated automation that reasons about your own data. Media server maintenance, smart home orchestration, data pipeline management—all coordinated by an agent that runs at home.
The MCP support is the unlock. Right now, if you want an agent to call a tool, you configure it explicitly. With MCP, the agent can browse available tools at runtime. That means less configuration and simpler workflows.
Getting Started
Docker is the fastest way:
docker run -d --name n8n -p 5678:5678 n8nio/n8n
Pair it with Ollama (for local LLM inference) and Open WebUI (for the LLM interface with MCP support). Use PostgreSQL to persist workflow history.
The Audience
The search volume for “n8n” sits around 9,600 queries per month, with low commercial intent ($7.43 CPC). That’s the audience: technologists and homelab builders who know what workflow automation is and want it self-hosted.
If you’ve been automating with Zapier or Make, N8N offers the same power on your own hardware. If you’ve been scripting workflows in Python, N8N’s GUI approach—now with Agents and MCP—is faster to iterate on. And if you just want Ollama to coordinate work across your lab, this is the tool to try.
For details, see the official N8N changelog and release documentation.