n8n + MCP: Build and Automate Anything Locally

Source: YouTube — WorldofAI, published 2025-04-20 Tools: n8n, MCP, n8n-nodes-mcp (community node), Brave Search MCP

Summary

WorldofAI demonstrates wiring MCP servers into n8n via the n8n-nodes-mcp community node (by nerding-io). The setup: install n8n via Docker with N8N_NODES_EXCLUDE_CUSTOM=false to allow community nodes, install n8n-nodes-mcp, then build a workflow with chat trigger → chat model → MCP client (list tools) → MCP client (execute tool). The demo connects the Brave Search MCP server as a tool, so an n8n AI agent can do real-time web search inside any workflow. The pattern generalizes: any MCP server (Perplexity, Filesystem, Postgres, custom) drops into n8n the same way.

Key facts

  • Community node: n8n-nodes-mcp from nerding-io
  • Required env: N8N_NODES_EXCLUDE_CUSTOM=false in the n8n container
  • Workflow shape: chat trigger → chat model → MCP client (list-tools) → MCP client (execute-tool)
  • MCP install path: Docker Desktop MCP CLI / npx pattern (e.g. npx @brave-ai/brave-search-mcp)
  • Demo: Brave Search MCP wired to an OpenAI-backed n8n agent — real-time search with citations
  • Model flexibility: works with OpenAI, Ollama local, OpenRouter, any chat-model n8n supports

Why it matters

This is the second n8n+MCP angle in the wiki, after the first-class Perplexity node from v1.98+. The two cover complementary patterns:

PatternWhen to use
First-class node (Perplexity v1.98+)When n8n has shipped a native node for the tool you need
Community MCP node (n8n-nodes-mcp)When n8n hasn’t shipped a native node, but an MCP server exists

The MCP path is the escape hatch: it gives n8n access to the entire MCP ecosystem (Brave, Filesystem, Postgres, Stagehand, Context7, Mem0 — anything wrapped as MCP) without waiting for n8n to ship a first-class node for each one. That makes n8n + MCP the wiki’s most general-purpose no-code agent runtime.

See Also