automation Official server available

n8n MCP Server

n8n MCP explained: n8n's built-in instance-level MCP server vs the czlonkowski/n8n-mcp repo. Let Claude build, fix and run your marketing workflows.

Pricing
free
GitHub stars
22,969
Last commit
2026-09-21
Page updated

Both are free. The community server is MIT licensed (hosted tier free up to 100 calls/day); n8n is free to self-host or paid on n8n Cloud.

There are two n8n MCP servers and they do different jobs. n8n itself ships an instance-level MCP server (Settings → Instance-level MCP) on both Cloud and self-hosted. Claude can search your workflows, run the ones you've exposed, and since n8n 2.13.0 it can create and edit workflows too. The community czlonkowski/n8n-mcp is more of a knowledge base: it gives Claude docs and property schemas for 2,791 nodes (833 core, 1,958 community) plus templates, so the workflows it writes actually validate.

Plenty of people run both. Use the built-in server to trigger and manage workflows, and add czlonkowski's when you want Claude to design new ones from scratch. Give the community server an n8n API key and it can create, update, test and deploy workflows as well.

A marketing example: "Build a workflow that checks our blog RSS feed every morning, writes a LinkedIn post and an X post for each new article with an AI node, and drops both into a Google Sheet for review." Claude picks the nodes, fills in parameters, validates the result and pushes it to your instance. More on the tool itself on our n8n page.

The czlonkowski README says it plainly: never let AI edit production workflows directly. Duplicate first, test on a dev instance, keep exports. On the built-in server, only published workflows with a webhook, form, schedule or chat trigger can be exposed, and you switch each one on yourself.

Official vs community n8n MCP

n8n instance-level MCP

Remote server URL: https://<your-n8n-domain>/mcp-server/http

The built-in server is the easy win if you already run n8n: OAuth, per-workflow access, and it works in Claude.ai and ChatGPT as well as coding tools. It doesn't carry deep knowledge of every node, which is where czlonkowski/n8n-mcp helps when Claude is designing new workflows. Use the built-in one to run and manage, add the community one to build.

What you can ask it

  • “Build an n8n workflow that pulls new Typeform responses, scores each lead with an AI node and posts hot leads to #sales in Slack”
  • “Run the Weekly SEO Report workflow and tell me what it returned”
  • “Open my Newsletter Signup workflow and add error handling so failed Mailchimp calls email me”
  • “Find a template that turns blog posts into social posts and deploy it to my instance as an inactive draft”

Features

  • Built-in server: expose chosen workflows to Claude and run them on demand
  • Built-in server: build and edit workflows and data tables (n8n 2.13.0+)
  • Built-in server: OAuth or API-key auth, with per-client access you can revoke
  • Community server: docs and property schemas for 2,791 n8n nodes
  • Community server: validate single nodes or whole workflows, with an autofix tool
  • Community server: search workflow templates and deploy them to your instance
  • Community server: create, update, test and version workflows through the n8n API

How to set up the n8n MCP Server

Works with: Claude, Claude Code, ChatGPT, Cursor, VS Code, Windsurf

Option A: n8n's built-in MCP server 1. In n8n, go to Settings → Instance-level MCP and click Enable MCP access (owner or admin only). 2. Click Connect, stay on the OAuth tab and pick your client. Claude.ai and ChatGPT get a one-click setup, or you copy the Server URL ending in /mcp-server/http. 3. For Claude Code with an API key from the API key tab:
claude mcp add --transport http n8n-mcp https://<your-n8n-domain>/mcp-server/http \
  --header "Authorization: Bearer <YOUR_N8N_MCP_TOKEN>"
4. Expose workflows one by one: open a workflow, then ⋯ → Settings → Available in MCP. Option B: czlonkowski/n8n-mcp (community) 1. Add this to claude_desktop_config.json. Drop the two N8N_ lines if you only want node docs and validation:
{
  "mcpServers": {
    "n8n-mcp": {
      "command": "npx",
      "args": [
        "n8n-mcp"
      ],
      "env": {
        "MCP_MODE": "stdio",
        "LOG_LEVEL": "error",
        "DISABLE_CONSOLE_OUTPUT": "true",
        "N8N_API_URL": "https://your-n8n-instance.com",
        "N8N_API_KEY": "your-api-key"
      }
    }
  }
}
2. Restart Claude Desktop. Cursor, Windsurf and Claude Code work too. 3. Don't want to run it? The hosted version at dashboard.n8n-mcp.com has a free tier of 100 tool calls a day.

Alternatives

Zapier MCP

Fully hosted with 9,000+ apps and nothing to maintain, but each tool call costs two Zapier tasks.

FAQ

Does n8n have an official MCP server?
Yes. n8n has a built-in instance-level MCP server on Cloud and self-hosted, found under Settings → Instance-level MCP. There's also an MCP Server Trigger node if you want to expose tools from a single workflow.
What's the difference between n8n's MCP and czlonkowski/n8n-mcp?
n8n's server runs and edits workflows on your instance. czlonkowski/n8n-mcp gives Claude documentation and validation for 2,791 nodes so it builds correct workflows, and can also manage them through the n8n API.
How do I connect n8n to Claude?
Enable MCP access in n8n settings, click Connect and choose Claude. Claude.ai gets a one-click setup; Claude Code uses claude mcp add with your /mcp-server/http URL.
Is it safe to let Claude edit my n8n workflows?
Not on production ones. Duplicate the workflow, test on a dev instance and keep exports, which is exactly what the czlonkowski README tells you to do.

Added: 2026-03-08 · Last reviewed: 2026-09-22