Install a node or give AI a skill to extend n8n

The two ways to extend n8n

n8n has two extension models, both covered here:

• Community Node: adds a node to the canvas for use in a workflow, such as Puppeteer for web capture or ElevenLabs for TTS. Every npm package named n8n-nodes-* belongs to this category. By 2026-08 there were more than 500 packages and 5,800+ skill-level units.
• AI Agent Skill / MCP Server: gives your AI (Claude, Cursor, or Windsurf) another capability. Connect n8n-mcp and ask it to build a workflow; AI calls the n8n API directly. This model teaches AI how to communicate with n8n and is separate from the workflow itself.

Both expand what you can automate. A community node gives n8n another capability; an MCP server removes a task from the person operating n8n.

How to install (the 3-minute version)

Install a community node

1. Open n8n Settings → Community Nodes

   Admin access is required. Self-installation is disabled by default in the managed service (n8n.woowtech.io); contact WoowTech. It is available on a local add-on or other self-hosted n8n instance.

2. Select Install a Community Node and paste the npm package name

   For example: n8n-nodes-mcp, n8n-nodes-puppeteer, @elevenlabs/n8n-nodes-elevenlabs. Accept the risks of third-party packages, then select Install.

3. Reload n8n (or wait for it to restart automatically)

   After installation, search for the node in the editor’s nodes panel. If it appears, installation succeeded. If not, check Executions → Errors; package incompatibility is a common cause.

4. Add the node and run it once

   The official template or repository README usually includes a minimal working example. Connect credentials and test it.

Install an MCP server (for your AI)

In Claude Code, Cursor, or Windsurf, add this to settings.json or the MCP config:

{
  "mcpServers": {
    "n8n-mcp": {
      "command": "npx",
      "args": ["-y", "[email protected]"],
      "env": {
        "N8N_API_URL": "https://n8n.example.com",
        "N8N_API_KEY": "<your API key, created under n8n Settings → n8n API>"
      }
    }
  }
}

Save and restart the AI client. With the pinned server release, tools such as n8n_create_workflow and n8n_list_workflows will appear. Prompt Library contains 44 copy-ready prompts.

Ten community nodes worth knowing (from an SMB angle)

From restyler/awesome-n8n, a list sorted by monthly npm downloads, these are the ten most useful picks for small and midsize businesses and internal teams, with the closest fits first:

Download counts are from restyler/awesome-n8n monthly-download snapshots dated 2026-01-20 and may be higher by 2026-08.

n8n MCP servers · Give your AI a workflow builder

An MCP server teaches your AI how to communicate with n8n. After connecting it, ask Claude or Cursor to build a workflow and AI uses the n8n REST API to create, activate, or debug it. These are the most active options:

npx -y [email protected]

npx @makafeli/n8n-workflow-builder

npm i -g @kernel.salacoste/n8n-workflow-builder

npx n8n-workflow-builder-mcp

Also listed: Jimmy974/n8n-workflow-builder, daniel-carreon/n8n-workflow-builder-mcp-server—similar forks with broadly comparable features for users who prefer a particular maintainer.

Six popular collections to browse

To browse rather than pick packages one at a time, start with these six collections:

# Index only; follow the links to install each original npm package
https://github.com/restyler/awesome-n8n

npmjs.com/search?q=keywords:n8n-community-node-package

n8n.io/integrations

n8n.io/workflows

community.n8n.io

github.com/topics/n8n-nodes

Compatibility notes

Whether a community node or MCP server runs in your environment usually depends on three things:

• n8n version: a node often pins a range of n8n minor versions. WoowTech’s n8n.woowtech.io is recorded in this repository as running v2.32.7 Community Edition; the public settings endpoint does not expose a version, so confirm it in the instance. Before installation, check peerDependencies.
• Node.js version: a node using newer syntax such as optional chaining or top-level await will fail when n8n runs on an older Node.js. The HAOS add-on defaults to Node 20 and is usually compatible.
• Additional service dependencies: Puppeteer, for example, needs a Chromium binary not included in the n8n Docker image. Add a Dockerfile or use the standalone image for n8n-nodes-puppeteer.

MCP client support alone is not enough: the server must use routes and features available in your n8n version. For the repository’s pinned n8n v2.32.7, use only the Public API operations documented for that version and omit MCP features whose server documentation requires a later n8n release. Recheck compatibility before upgrading either side.

Safety rules

• Read the README before installing: not every package clearly documents the credentials it needs or the requests it sends. Ask Claude to review it if necessary.
• Pin versions in production: in package.json, use n8n-nodes-xxx: 1.2.3 rather than ^1.2.3 to prevent an unnoticed upgrade from introducing an upstream regression.
• Apply least privilege to MCP: on Enterprise, create a dedicated key with only the required scopes and project access. Community/non-Enterprise keys have the creator’s full account capabilities; they cannot be made read-only or tag-scoped. Use a dedicated least-privileged user/project where available, disable write tools, or keep MCP away from production. Do not use an owner/admin key unless an owner/admin-only operation, such as credential listing, is required.
• Review contributors: a node with only 1 contributor may stop receiving updates if that person leaves. Packages maintained by an official or commercial organization are safer for production.
• Run an audit after installation: the pinned server provides n8n_audit_instance, backed in part by the documented POST /api/v1/audit route. Review the report after installing a node.

SKILL.md spec at a glance (for AI agent skills)

If you want to write an AI Agent Skill for Claude or Cursor, Anthropic’s SKILL.md format is the current de facto industry standard. The smallest structure is:

n8n-workflow-review/
└── SKILL.md          ← the only required file

# SKILL.md contents
---
name: n8n-workflow-review
description: When a user provides n8n workflow JSON, check its structure—
  find dangling connections, missing credentials, and clear performance problems.
  Also trigger when the user asks “Does this workflow have a problem?” or “Review this.”
---

# n8n Workflow Review

(Instructions for the agent, written as if onboarding a colleague)

Keep SKILL.md concise. Put supporting files under references/, scripts/, assets/ and read them only when needed to avoid filling the AI context.

FAQ

Take this handbook with you

This handbook is a self-contained single HTML file that works offline. Collection data ages, so before an important decision return to the Resource Hub for the online version.

Download this handbook as HTML Back to Resource Hub