n8n MCP Claude Integration: Control Workflows Directly from AI

Key Takeaways

• Version Requirement: You must be on the latest version of n8n (Cloud or Self-hosted) to access MCP features.
• One-Time Setup: Connect n8n as an MCP server within Claude Desktop once, and it persists.
• Granular Permission: You must manually toggle “Available in MCP” inside each specific n8n workflow settings to make it visible to Claude.
• Direct Execution: You can trigger, debug, and pass data to workflows using natural language prompts in Claude.

If you are tired of constantly switching tabs between your AI chat interface and your automation dashboard, the new Model Context Protocol (MCP) integration is a game-changer. It allows Claude to “see” your n8n instance and execute workflows directly.

I’ve been using n8n for years, and this feature bridges the gap between static automation and agentic AI. Instead of just building pipelines, you are now giving your AI agent hands to do the work. Here is how to set up the n8n MCP server and connect it to Claude in under two minutes.

Prerequisites and System Updates

Do I need to update n8n to use MCP?
Yes, this is a relatively new feature, so running the latest build is mandatory.

Before attempting the connection, check your current version.

1. Navigate to Settings > Health in your n8n dashboard.
2. Check how many versions behind you are.

If you are on n8n Cloud, the update can be done via the UI. If you are self-hosting on a VPS (Docker), you need to pull the latest image via your terminal. Don’t skip this; the “Instance Level MCP” option simply won’t appear in older versions.

Configuring the n8n MCP Server

Once your system is up to date, we need to grab the connection string.

1. Open your n8n dashboard.
2. Go to Settings.
3. Select Instance Level MCP (sometimes labeled under “Connectors” or “New Features” depending on your specific build).
4. Locate the Server URL.
5. Click the copy button.

PRO TIP:

Keep this URL secure. It acts as a gateway to your automation engine. If you regenerate the URL, you will need to re-authenticate inside Claude.

Connecting to Claude Desktop

Now, we move to the client side. Open your Claude Desktop application.

1. Click the + (Plus) button or go to Settings.
2. Select Add Connector.
3. Choose the Web tab (since we are connecting via a URL, not a local file).
4. Paste the n8n Server URL you just copied.
5. Click Continue.

Claude will present a permission modal asking if you want to allow it to access your n8n instance. Click Allow.

If successful, you will see a confirmation, and the connector status will turn active. You can toggle this connection on or off at any time via the Claude connector settings if you need to revoke access temporarily.

Exposing Workflows to the AI

Connecting the server is only step one. By default, for security reasons, Claude cannot see any of your workflows. You must explicitly grant access to specific automations.

Let’s use a standard “Form to Google Sheet” workflow as an example. This workflow accepts data (Name, Email, Request) and appends it to a spreadsheet.

To make this controllable by AI:

1. Open the specific workflow in the n8n editor.
2. Click the Menu (three dots) or Settings icon in the corner.
3. Find the toggle labeled Available in MCP.
4. Switch it to ON.
5. Save the workflow.

If you forget to save, the MCP server won’t register the change, and Claude will hallucinate that it can’t find the tool.

Executing Automation via Prompts

Now for the actual test. We will ask Claude to find the workflow and run it with sample data.

The Prompt:
Input a prompt similar to this into Claude:

“Search the n8n MCP server for the ‘Form to Google Sheet’ workflow and execute it with the following data: Name: Test User, Email: [email protected], Request: Test submission from Cloud.”

What happens in the background:

1. Discovery: Claude queries the n8n MCP tool definition list.
2. Matching: It finds the workflow ID matching your description.
3. Execution: It sends a JSON payload to the n8n webhook/trigger.
4. Response: n8n runs the logic (writing to Google Sheets) and returns the execution output to Claude.

In my testing, Claude immediately confirms: “I have found the workflow and started execution.”

To verify, go back to n8n and look at the Executions tab. You should see a successful run. Open your Google Sheet, and the new row with “Test User” should be populated instantly.

PRO TIP:

Use descriptive names for your workflows. If you name a workflow “Untitled 1,” Claude won’t know when to use it. Name it “Add Lead to CRM” or “Scrape Website Data” so the AI allows for semantic matching.

Frequently Asked Questions

• Q: Can I use this with the web version of Claude.ai?
  • A: Currently, MCP support is primarily designed for the Claude Desktop app and compatible IDEs (like Cursor or Windsurf). Browser-based support is evolving, but the Desktop app is the most stable method right now.
• Q: Is it safe to expose my n8n instance?
  • A: Yes, but you must be intentional. The “Instance Level MCP” only connects the server. The AI cannot run a workflow unless you manually toggle the “Available in MCP” setting inside that specific workflow. This prevents the AI from accidentally triggering critical or destructive pipelines.
• Q: What if Claude says “Tool not found”?
  • A: This usually happens for two reasons:
    • You didn’t save the workflow after toggling “Available in MCP”.
    • The workflow name is too ambiguous. Try renaming the workflow to something very specific and retry the prompt.