🚀 I Built an MCP Server That Makes Claude an n8n Expert - Here's How It Changed Everything

:rocket: Claude Desktop Can Now Build Perfect n8n Workflows - Thanks to MCP

Hey n8n community! :waving_hand:

TL;DR: I built an MCP server that gives Claude Desktop complete, up-to-date knowledge of all 525 n8n nodes. That changes it from a helpful tool into real AI n8n buildier. Now it can build workflows almost perfectly on the first try. What used to take 45 minutes with errors now takes 3 minutes with zero mistakes.

The Problem

Remember when Claude would guess node names wrong and mix up properties?

Me: "Create a Slack webhook workflow"
Claude: "Use slackNode with message property..."
Me: "It's 'slack' with 'text' property..."
Claude: "Oh, webhookTrigger then?"
Me: "Just 'webhook'..."

45 painful minutes later, maybe you’d have a working workflow.

The Solution: n8n-MCP

I created a MCP service, that serves as always updated n8n documentation knowledge base, with examples and validation tools. Now Claude has direct access to:

  • :white_check_mark: All 525 n8n nodes with complete documentation
  • :white_check_mark: Every property and operation (no more guessing!)
  • :white_check_mark: Working examples for common tasks
  • :white_check_mark: Real-time validation before deployment

Result: 45 minutes → 3 minutes, 6 errors → 0 errors

Quick Installation (5 Minutes)

Prerequisites: Docker Desktop installed on your system

Pull the Docker image

docker pull ghcr.io/czlonkowski/n8n-mcp:latest

Add this to your Claude Desktop config:

{
  "mcpServers": {
    "n8n-mcp": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "-e", "MCP_MODE=stdio",
        "-e", "LOG_LEVEL=error",
        "-e", "DISABLE_CONSOLE_OUTPUT=true",
        "ghcr.io/czlonkowski/n8n-mcp:latest"
      ]
    }
  }
}

Config locations:

  • Mac: ~/Library/Application Support/Claude/claude_desktop_config.json
  • Windows: %APPDATA%\Claude\claude_desktop_config.json

Restart Claude Desktop and you’re done! :tada:

What You Can Do Now

Ask Claude to:

  • “Build a workflow that monitors RSS feeds and posts to Discord”
  • “Create an API endpoint that validates data and saves to Postgres”
  • “Set up a daily report that pulls from multiple sources”

Claude will deliver working JSON you can paste directly into n8n.

More Options & Details

:link: GitHub: github.com/czlonkowski/n8n-mcp

  • Local installation options
  • Full documentation
  • MIT licensed (free forever)

Feedback Welcome!

What workflow challenges do you face? What would make Claude even better at n8n? Drop a comment or open an issue on GitHub.

Let’s make n8n + AI workflow creation delightful! :star:


P.S. - Claude even discovered features I didn’t know about, like Google Sheets’ built-in duplicate detection. Sometimes the AI teaches the human! :grinning_face_with_smiling_eyes:

12 Likes

Sounds nice! But i tried it with several use cases, including your “Build a workflow that monitors RSS feeds and posts to Discord” use case.

Hi freddy! I am glad you checked it out! The key here is to use Project and have system prompt with instructions (included in the github repo) - I also included templates in the tools, so that should improve the outputs. Also, be sure to use Opus 4 with extended thinking. Let me know if it helped!

Hi @romuald , thanks for sharing, I’m a beginner with n8n, I was stuck with an email copywriting workflow that should send request to Anthropic by Batches. I gave my json to claude with your n8n-mcp. It optimized my workflow and made it work. Really nice :slight_smile:

2 Likes

New version just rolled out with advanced node / workflow validation features - problems should be now reduced significantly! :slight_smile: Be sure to use the latest image

Dang, this is super impressive!!!
Did you write it yourself?
HOW?
Teach me.

BIG THANK YOU!

1 Like

I am facing the following error, working with n8n Version 1.100.1 and n8n running in docker without any proxy all works fine (telegram too) now when I try to configure a mcp server I get the correct URL and checking with cursor it works fine. The issue is with the MCP client, I can´t see the MCP server. Any help please?

Hi! connecting to AI Agents in n8n does not work yet - here you can follow the issue resolution

Hey thanks for your great work!

I’m running into an issue where i tried with multiple models from openai and from anthropic
tried multiple scenarios
but 90% of the time its creating flows with faulty nodes

Not sure where to address this issue - but if the Claude hallucinates something, even though it’s supposed to “validate” the workflow nodes, the n8n UI can’t load the JSON workflow. As result, yo get an “empty” project, and this error " propertyValues[itemName] is not iterable", or anything alike:

I understand that Claude might hallucinate stuff, but it’s impossible to work around this, when the n8n UI doesn’t show even which node and what property is invalid.

Any ideas on how to resolve this?

1 Like

Hey @romuald I’m getting this error, can you help me debug? Maybe I’m missing something, I’m not super technical — thanks so much!

@romuald ,

Thank you for creating such a valuable tool.

I’ve got it set up and working, BUT I keep running into the same error:
“Claude hit the maximum length for this conversation. Please start a new conversation to continue chatting with Claude.”

That was Claude’s response to this short prompt:
“Please examine the n8n workflow with the ID of iVnYHYqUyetpg8w1 and guide me on the best way to set up this workflow and identify any potential problems I should address with it.”

The workflow contained only two nodes and was a starter workflow for error monitoring of my n8n main workflow. Claude provided about a page of suggestions and then indicated the maximum length error.

I got the same error response with both Opus 4.0 (extended thinking = on) and Sonnet 4.0. I know the error ISN’T from your MCP server. I am the paid Anthropic “Max” plan at $100/month.

If it matters, my project knowledge is only about 6% full, and my active integrations look like this:

and

I’m wondering if I should expect Claude to be able to use the n8n-MCP server for larger conversations than I’m currently experiencing. Perhaps there is an install step I missed or an add-on I forgot to enable.

I appreciate any feedback you might have.

Best Regards,

Bill

I had the same problem and accessed the MCP with VS Code. Since then no problems anymore.

Dears,

The community response to this MCP server is nothing short of absolutely amazing!

If you encounter issues, please report an issue on github

I apologize, but I don’t have the bandwith to help troubleshoot individual issues. Me and early contributors are working hard to make the tool better every day, so remember to update often.

If you are able to help make this tool better, consider contributing as well!

3 Likes

@romuald @Andrea_Groblin any feedback would be appreciated, thanks!

I followed your instructions to install the docker image, and made the config changes, but the MCP does not run in claude desktop, and i can’t add it as a custom connector. What obvious thing have i missed? I’ve spent a couple of hours working with Claude and Gemini to troubleshoot my install.

Use claude desktop to help you. just paste in the errors that you are getting.

yup, worked thru it with various LLMs for a few hours, and tried all that was suggest, but no luck