Keep n8n Workflows Running
With Your AI
Connect Claude, ChatGPT, Cursor, and 20+ AI tools to the n8n you already run. They read failed executions, patch the broken node, test the fix, and build what's next — across every instance you manage.
From the founder of n8n
“I love the work you did. You really empowered a lot of people. What you built with the MCP is really amazing.”

Jan Oberhauser
Founder & CEO, n8n
Trusted by teams at
Keep It Running
Workflows Break. Your AI Can Fix Them.
n8n-mcp doesn't ship an agent — it's the connector that gives the AI you already use the hands for production work: honest validation, testing, real execution logs, surgical edits, and version history with rollback. Where the Public API stops, n8n-mcp talks to n8n's native MCP server: pinned-data test runs, native version history, live dropdowns, n8n Agents. The intelligence is yours; the tools are ours.
An agent loop is only as good as its stop signal. n8n-mcp fails loudly on unknown nodes, invalid versions, and missing required fields — so when your AI says "valid", it means it.
Private by design
Your workflows, executions, and credentials stay in your n8n instance — the AI reads only what you connect. Where the platform learns from usage, it stores workflow patterns, not your data, and every contribution is screened by self-hosted AI models before ingestion — nothing is sent to external services.
One Connection
Every n8n You Manage. One Connection.
Production and staging. Your own instance and your clients'. Register them once, and the AI you already use works across all of them from a single connection.
You ask
“Run a health check on each of my client instances and tell me which one is down.”
One connection reaches
- proddefaulthealthy
- staginghealthy
- client-acmedown
- client-nordichealthy
Name each instance once
Register prod, staging, and client instances in the dashboard. Say “deploy to staging” and your agent targets it — no second account, no swapping API keys.
Target per call
Any call can name its instance, so parallel agents can work on different instances at once. A name that doesn't match is refused, not guessed.
Keys stay server-side
n8n API keys are stored encrypted and never returned through MCP. Your agent sees instance names and URLs, nothing more.
Included on every plan of the hosted service. n8n's built-in MCP server connects to one instance at a time.
How multi-instance mode worksn8n's Native MCP
Already Using n8n's Built-In MCP? Keep It.
n8n ships its own MCP server, and it covers a lot. n8n-mcp is built to sit on top of it, not to replace it. Here is how to tell which you need.
The built-in server is enough when
- You work with a single n8n instance
- That instance runs a recent n8n 2.x
- You build and edit workflows on that one instance
- You set up credentials yourself in the n8n UI
Add n8n-mcp when
- You run more than one instance — prod and staging, or your clients'
- Your AI should create and update credentials, not only list them
- You're on an older n8n — 1.x or early 2.x — that predates most built-in MCP tools
- You want validation that fails loudly, autofix, execution error analysis, and a security audit
Or use both through one connection
Add your instance-level MCP token next to your API key and n8n-mcp routes the native-only features through n8n's own server: test runs for any trigger with pinned data, n8n's version history with diff and rollback, live dropdown values, and n8n Agents. Your AI still sees one MCP server.
Choose Your Way
Two Ways to Run n8n-MCP
The zero-setup hosted service, or the open-source server on your own machine. Same tools for your AI — pick the way that fits.
Hosted
The managed n8n-MCP service. One-click OAuth from claude.ai, ChatGPT, and your IDE, node documentation synced with every n8n release, n8n's native MCP features, and multiple n8n instances from a single account — no server to run.
Sign up
Create a free account on the dashboard
Connect
One-click OAuth for your AI tool, then your n8n API key — plus the instance-level MCP token for native features
Operate & build
Your AI tool can now debug, edit, test, and build workflows on your n8n
What your AI gets
Self-Hosted
The open-source MIT server, running on your machine. The same toolkit as hosted, on your infrastructure — including n8n instances the hosted service can't reach.
# Quick start
npx n8n-mcp
# Or with Docker
docker run -i ghcr.io/czlonkowski/n8n-mcp
# Optional: n8n's native MCP (n8n 2.34+)
N8N_MCP_ACCESS_TOKEN=...
- Works with localhost n8n instances
- No quotas, no account needed
- MIT licensed, free forever
- Your infrastructure, your data
The EvolutionFrom Frustration to Flow
This is what AI-built workflows look like today. Hit run.
Copy-Pasting JSON
Unknown property "webhookId"AI generates workflow JSON that you copy into n8n. It looks right, but fails with cryptic errors because of hallucinated properties.
Screenshotting Workflows
Input: screenshot.png (no context)Want AI to help improve your workflow? Screenshot it, paste it back, explain the context. Every. Single. Time.
Guessed IDs, Stale Options
Channel "C0ABC123" not foundAI invents Slack channel IDs, sheet names, and model names it has never seen — against node schemas that are months out of date.
Blind Debugging
Execution log: not visible to AIWorkflow failed? AI can't see the execution logs. You're stuck copy-pasting error messages and hoping for the best.
Copy-Pasting JSON
AI generates workflow JSON that you copy into n8n. It looks right, but fails with cryptic errors because of hallucinated properties.
Direct Deployment
AI creates workflows directly in your n8n instance. No copy-paste, no import errors. Just working automation.
n8n_create_workflow · n8n_update_partial_workflowCommunity
What builders say.
Real experiences from people building with n8n-mcp.
32 testimonials. Scroll horizontally, drag, or use the arrow keys to explore. Automatic movement pauses while reading or interacting. Use the pause button to stop it.
For me this is such a HUGE enabler, allowing someone like me coming from the operations side of things, to build workflows based on actual processes and doing my own POCs, without having to involve other figures at early stages.

Then I found n8n-mcp and azure-mcp… and solved everything in 10 minutes. This is a game changer for so many people.

your n8n-mcp was the primary reason I could start building with n8n and gradually become a power user :)

From a new user of the n8n-mcp.com platform - has been super useful so far, lots to learn and understand but definitely a time saver when configuring workflows!

Love it! Especially because it makes it way easier to maintain complex automations compared to ones created using other AI agents

It saves me a ton of time every week.

I recently started following your work and I’m one of the people using it to build workflows, it’s honestly improved the development time massively. … the tool amplifies your thinking rather than replacing it.

I built a similar flow in n8n for a lead magnet of mine too and 100% was thanks to the n8n MCP so thank you

after my initial problems using this MCP I tried it again recently with more intelligent models and was very pleased with the results. … Helped a client migrate some pretty complex flows from Relay.app and it worked very well with both Opus 5 and GPT 5.6 Sol.

What a great feature. I use n8n MCP every day. Thank you.

Thank you for your work. It's my most used MCP. No more control C, control V!

Thanks for a great piece of code, use it every day in my flow!

I just created my n8n agent in VS Code, then installed your n8n skills and n8n MCP. Oh my god, it was amazing! Thank you!

I’ve been using n8n-mcp for a few months now, and I really appreciate how much it streamlines my work.

Romuald, I use it since about 10 month and it’s still one of the best solution you have build so far. Thank you 🙏 for your efforts.

Truly amazing work! I’ve been using n8n-mcp and it works great; very excited to try out skills!

I need to let you know that I'm still getting better results with your MCP than the newly integrated N8N chat!! Amazing!

I have tied in the n8n-MCP from my docker instance to my perplexity pro app. … Works amazingly well!

I tested the n8n-mcp + n8n-skills integration using Claude code with your repo and it performed very well at creating n8n workflows using Claude code as the AI engine of this system.

still crazy I'm only one out of 3100 guys using your tool actively everyday. :D

One of the best MCPs out there, and coupled with n8n Skills it’s unbeatable.

After testing your n8n MCP, I was honestly blown away by the idea of automating… automation itself.

Superb package by the way and still my goto one for any n8n effort.

I can't describe the feeling of first connecting Claude Code to n8n...let's just say I let out the type of laugh you see in movies from super villains. Cheers for that.

FWIW, your work has helped us at n8n quite a lot.

I have to thank you! Great to have people like you with us who drive that much impact.

Thank you for your continued work providing such an essential and valuable agentic tool to the n8n community.

Thank you for that, I use it daily!

Open source • MIT Licensed • Active community
Simple Pricing
Start Free, Upgrade Anytime
Free is enough to try it on a real workflow. Supporter is for daily work on production n8n — and it funds the open-source server.
* Fair-use cap of 5,000 tool calls per day. Let us know if you need more.
Running n8n for clients or a team? Tell us what you'd need from a plan built for many instances.
FAQ
Got Questions?
Real questions from real users.
The Next Time a Workflow Breaks, Ask Your AI
Connect the n8n you already run. Your AI reads the execution, finds the cause, and patches the node. Free to start, no credit card needed.
Need it built for you? Work with AiAdvisors — automation audits, builds, and operations by the team behind n8n-mcp.

















