Create workflows via MCP 🎉 Now in Beta!

Announcements
1

Up until now, n8n’s instance-level MCP server could only execute workflows from external tools like Claude Code, Cursor, and others.

We’ve just added two new capabilities to the n8n MCP server, now in beta:

  1. Create new workflows

  2. Update existing workflows

This is now live in the 2.14.0 beta and available for everyone - cloud, self-hosted community, and enterprise versions.

This has been one of our most consistent requests, and we get why. Many of you have told us you want to build and modify workflows from within your own AI tooling without switching back to the editor for every change.

It’s also part of a larger effort that’s become a core focus for us at the product level: making n8n work better alongside other tools. Whatever tool you’re building with, n8n can be a first-class participant, not something siloed in its own editor. For example, using n8n together with Claude Code, Cursor, or Codex.

We’ve already seen this happening in our community, some of which were ahead of us.

On that note, czlonkowski/n8n-mcp has been helping people build n8n workflows with an MCP server for a while now, with 14k+ stars on GitHub. We appreciate the work that’s gone into it and encourage you to check it out if you haven’t.

We also want to make sure the external tools people use to build with n8n (the MCP clients) have as complete an understanding of n8n as possible. That means access to internals that aren’t exposed through the public API, which is why we’re building this natively.

The public beta is available now, with a full announcement to follow, but we wanted to share it with our community members first.

Here’s how to connect to the n8n MCP server - please note this page has not been updated to reflect the new MCP capabilities.

Let us know what questions or comments you have so we can help as you try it out.

Thank you!
Ophir & the rest of the n8n team

26 Likes
3

This is great to see. Native workflow creation/update via MCP is a natural evolution and something the community clearly needs.

On the topic of making n8n work better alongside other tools, I’ve been working on a complementary approach: n8n-cli, a command-line interface for the n8n REST API designed specifically for AI agents.
https://github.com/MauricioPerera/n8n-cli
The key difference from MCP-based approaches: it’s transport-agnostic. Any AI agent, script, or CI/CD pipeline that can run a shell command gets full n8n access, no MCP client required.

A few highlights that might be relevant to this discussion:

  • Node catalog with search: agents can discover available nodes and inspect their full parameter schemas before composing workflows (nodes search "send email" → nodes get n8n-nodes-base.gmail)

  • AI skill system: built-in commands like meta workflow-template, meta credential-guide, and meta ai-recipes that teach the AI the exact JSON structure, credential mapping, and connection types (including AI node types like ai_languageModel, ai_memory, ai_tool)

  • Multi-instance profiles: manage local, staging, and production from one CLI

  • Pure JSON I/O: every stdout is valid JSON, every error is structured JSON on stderr

The node catalog piece is something I’d love to see in the native MCP server too. One of the biggest challenges I found building my earlier project (n8n-mcp-claude) was that AI agents need to understand node parameters before they can compose correct workflow JSON. Without that discovery step, they hallucinate parameter names.

Would be happy to share findings or collaborate on making the native MCP’s node awareness richer. The CLI and MCP approaches complement each other well: MCP for tight IDE integrations, CLI for everything else.

Congrats on the 2.14.0 beta! :tada:

2 Likes
4

This is so awesome to see! I think n8n integrates especially well with Claude (At Synta, this is the number one client our customers use for connecting to Synta MCP to make n8n workflows) which seems to be the best at making and editing n8n workflows, but still needs guidance and the correct tools to really steer it. I think when you combine the ability to create and test it, you can get that self-healing loop!

Our team at Synta (the best n8n MCP and builder/editor for n8n workflows) have been building an n8n MCP that used by over 1000 businesses, and it’s been so amazing to see how n8n has progressed and developed over time with their native MCP as well, and I think a collaboration would be super interesting! We have a lot of insights and ideas on how n8n can improve and adjust their offering going forward, and with your release of the MCP, it could be a great combination!

1 Like
5

This is a big step forward :slightly_smiling_face:

6

Does this fix the Webhook registration bug via api?

7

what bug is that? do you mean auto-injecting a webhook id or?

8

This one: Deployment + Activation via API does not register webhook path · Issue #21614 · n8n-io/n8n · GitHub
But seems to be fixed in 2.14.0: fix(core): Assign webhook ID to API-created webhook nodes by ivov · Pull Request #27161 · n8n-io/n8n · GitHub

9

@Ophir_Prusak Thanks for the shoutout for n8n-mcp.com - means a lot coming from the team.

This is great news honestly. Native access to internals I’ve been working around from the outside will make everyone’s life easier, mine included.

n8n-MCP isn’t going anywhere though. The 104K workflow patterns, the AI generation models, the template search — that’s the stuff I’ll keep pushing on. Complements what you’re building here nicely.

Congrats on shipping it :rocket:

P.S @freddy-schuetz, if you build through n8n-mcp, webhook paths are automatically ingested offsetting this bug

10

@Ophir_Prusak Congrats on this huge release! :rocket:

And thanks for shouting out @Romuald_Czlonkowski community tool. It’s always great to see the n8n team highlighting the builders in the ecosystem.

Building on this AI-driven momentum, I wanted to share how we’ve been tackling the automation challenge from a different architectural angle with the n8n-as-code framework.

While the native MCP is brilliant for direct AI-to-canvas interactions via API, we chose to bring the environment awareness directly into the IDE (like Cursor) using local workspace skills. We’ve actually found this local-context approach to be incredibly fast and highly token-efficient, especially when generating complex workflows.

Combine that local AI awareness with strict TypeScript generation (instead of JSON), and you get a setup that completely eliminates hallucinations while natively supporting GitOps and CI/CD.

It’s amazing to see the ecosystem mature like this. The native MCP is a massive leap forward for the platform, and tools like n8n-as-code will keep pushing the boundaries on the Developer Experience (DX) side. Exciting times ahead!

11

That’s a great addition to the MCP server!

I would love to see the ability to retrieve details and update any workflows without having to activate MCP for them OR adding the subworkflow trigger as a valid option.

We have many subworkflows and adding manual triggers to them just to include them on the instance MCP seems like a needless overhead.