n8n returns with a NEW Tool (and SUPERCHARGES Claude Code)
CChase AI
Computing/SoftwareSmall Business/StartupsInternet Technology
Transcript
00:00:00it wasn't so long ago when N8N was everybody's favorite no code AI automation tool.
00:00:05Well, that was until other tools like ClodCode and Codex
00:00:09begin to completely overshadow it.
00:00:11But just this week, N8N released a brand new tool
00:00:14which may give you a reason to bring it back into the fold.
00:00:17So today we're going to dive into it
00:00:19and I'm going to let you know if N8N is worth your time again.
00:00:22Now, the tool I'm talking about is the brand new N8N MCP server,
00:00:26which was built with agentic coding tools like ClodCode and Codex in mind.
00:00:30Now, I think this is a big deal for a few reasons.
00:00:33The first is we haven't had a tool like this before.
00:00:37Well, sort of.
00:00:38We've had some Band-Aid fixes like Lonkowski's N8N MCP server
00:00:42and some other hacky things, but they were just that. They were hacky.
00:00:46They operated under the assumption that we could just load up
00:00:49the context window of something like ClodCode or Codex with,
00:00:53you know, a billion lines of documentation
00:00:56explaining here's what you should do, here's what you shouldn't.
00:00:58And hoping that because we shoved all that context in there,
00:01:02we could get JSON that actually worked.
00:01:04The other solution was just giant skill files that were just lines
00:01:08and lines and lines of markdown, again, trying to force ClodCode
00:01:11to create JSON in a way that worked inside of N8N.
00:01:14But this official MCP is a little bit different and a little more sophisticated
00:01:19because it's operating on TypeScript, not necessarily JSON.
00:01:24And what that means is we are able to validate the workflow
00:01:27and actually compile the code we're creating before
00:01:30it gets turned into a JSON file and populates inside of N8N.
00:01:34AKA, when I give ClodCode a prompt using this new N8N MCP server,
00:01:39we have some guardrails in place to make sure what we're creating actually works.
00:01:44So, for example, if I give ClodCode a prompt saying create me
00:01:50this simple weather automation that emails me every morning.
00:01:54First thing it's going to do is it's going to parse the intent and realize, OK,
00:01:58we're creating this very simple automation. I need three nodes.
00:02:01Well, it's then going to hit up that MCP server and get the node types.
00:02:06From there, it's then going to write code in TypeScript.
00:02:10That code is going to get sent to the MCP server and validated and parsed.
00:02:15This is a big difference.
00:02:16This is a huge difference between what we've done in previous MCP iterations.
00:02:20After that, it is then converted into JSON,
00:02:24where it then automatically is populated inside of your instance.
00:02:28One of the members of any team broke this down
00:02:31in a little bit more detail in this LinkedIn post, talking about this TypeScript
00:02:35thing to represent your workflow, not JSON and saying,
00:02:38when you ask the LLM to produce raw JSON for a workflow,
00:02:41it's guessing at the structure with no guardrails.
00:02:43That's what we've been doing up until now.
00:02:45But with this new server, TypeScript gives you type checking and compilation
00:02:50before anything touches your instance.
00:02:52The model has to produce something that actually compiles,
00:02:55which filters out a ton of errors.
00:02:58So you take all of that together.
00:02:59And this new and MCP server is the single best way to create
00:03:04any of the automations via coding agents like ClodCode.
00:03:07And nothing's ever really been close.
00:03:09Now, the second reason this MCP server matters
00:03:12is because I think N8N should still be part of your stack.
00:03:15This isn't 2024. This isn't 2025.
00:03:18I don't think N8N has as big of a part to play anymore.
00:03:21It's more niche.
00:03:23But if you're someone who works in the AI agency space, for example,
00:03:26you know the value of being able to create these sort of automations
00:03:30that are visual and that you can hand off to a client who isn't
00:03:34particularly technical, but for whatever reason, still needs to get hands
00:03:38on with this and telling them, oh, it's in a GitHub repo somewhere.
00:03:42Just spin up your own ClodCode instance is a bridge too far.
00:03:45In those cases, N8N does have real value, and it is kind of a pain
00:03:50in the butt to use ClodCode to gen up the JSON for you.
00:03:54It's kind of janky.
00:03:55But now it's kind of not.
00:03:57And because that is the niche that I think N8N operates in, the sort of automations
00:04:01you need to build with N8N don't have to be these wildly complicated things.
00:04:05If something's wildly complicated, well, I'm just going to use normal code,
00:04:08quote unquote, inside of ClodCode and go down that route.
00:04:11But if it needs to be pretty simple, someone who's non-technical
00:04:15needs to have like eyes on and hands on.
00:04:17Well, shoot, I can knock these out so fast now. It's so easy.
00:04:20Now, in terms of actually installing this thing, again, very simple.
00:04:24You need to make sure your N8N instance is up to date, first and foremost.
00:04:27Then you're going to go down here, you're going to go to settings.
00:04:32And you're going to go to instance level MCP.
00:04:34You can do this self hosted or on the cloud.
00:04:37Once you get to instance level MCP,
00:04:39what you're going to do is you and make sure this is set to enabled.
00:04:41You have the option here to enable workflows.
00:04:44If you enable a workflow, that's for workflows that already exist.
00:04:47So if you want to use this on workflows that you've already created,
00:04:50you have to go in here and enable them.
00:04:52But if you're just creating something new from scratch,
00:04:54you don't have to do anything.
00:04:57The other thing you need to do is hit connection details.
00:05:00You're going to want to do access token.
00:05:01If you're working in something like cloud code,
00:05:03then you just need to connect cloud code in this MCP server.
00:05:06So you're going to feed it the URL, the access token and the configuration JSON.
00:05:10If you want to test this out quickly, you can just drop the raw access
00:05:15token into the chat window, but understand that isn't best for security.
00:05:18You'll want to rotate it, set it as an environment variable.
00:05:21I'm not going to go through that step by step here for time.
00:05:23Just understand you need to do that.
00:05:24And cloud code will walk you through exactly the steps needed to complete that.
00:05:28But to get it up and running, make sure it actually works.
00:05:31Just copy paste all the stuff in there and you'll be good to go.
00:05:33Now, once you've copied and paste out the server URL, your access
00:05:36token in the JSON config into cloud code and said, hey, set up this MCP server for me.
00:05:41You're going to have to exit cloud code, start it back up and then just do forward
00:05:45slash MCP and you will see an end MCP now connected.
00:05:49If you don't see that,
00:05:51you either didn't reset it or you did something wrong.
00:05:54So just try again.
00:05:56And now to use it, all you have to do is just talk in natural language
00:06:00inside of cloud code, because once you set up that MCP, it now has a connection
00:06:04to any end itself, it will not only generate the code,
00:06:06it like actually builds it inside of your instance.
00:06:09There's no copy pasting back and forth.
00:06:10So if I say something like use the end MCP to build me a workflow
00:06:15that fires daily at nine a.m.,
00:06:16fetches Toronto weather and emails me the forecast.
00:06:19It's just going to go ahead and do it.
00:06:22So you can see here it's got the SDK and the node list.
00:06:25So it's figuring out which nodes it actually needs to use.
00:06:27It's written the workflow.
00:06:30Make sure it's valid.
00:06:31Now it's creating it.
00:06:33And then it's calling on the end of MCP
00:06:35to actually populate the workflow inside of my instance says it's complete.
00:06:40And we can see right here, Toronto daily weather email.
00:06:43And there we go.
00:06:46All right, if I click in here, cool, it has my email, has the message.
00:06:50Everything is already mapped by execute workflow.
00:06:55And I check my email.
00:06:56There it is now, obviously, extremely simple automation example.
00:07:01But that's how it works.
00:07:02Very, very simple.
00:07:03Now let's ask for something slightly more complicated.
00:07:06So I sit on a new canvas.
00:07:09I want us to create a newsletter automation that runs every morning at 10 a.m..
00:07:13I wanted to grab trending news across various RSS feeds.
00:07:17I'm letting it figure out what RSS feeds those should be,
00:07:20because I don't want to use an API key.
00:07:23I wanted to then pull it into an AI system.
00:07:25I wanted to use GPT five to summarize it and then email it off to me.
00:07:29So this time I'm asking for it to create a newsletter automation.
00:07:34Not extremely complicated, but I have had people pay for that.
00:07:38And for client work, for example, again, I think any niche
00:07:42these days is not the ultra complicated stuff.
00:07:44I could have it do examples of super ultra complicated stuff.
00:07:48But at that point, it's like, OK, why are we using it for that?
00:07:52Again, niche, we don't need to use any of them for everything,
00:07:56but if we are going to use it, I think this MCP is great.
00:07:58All right. So it finished up the newsletter automation.
00:07:59So let's take a look.
00:08:01See the new one right here.
00:08:05And there we go. It even looks pretty so at the trigger,
00:08:08it's pulling three different RSS feeds, merges them,
00:08:12filters them over the last 24 hours, aggregates them.
00:08:15Throws them in here.
00:08:18Here's sort of the system prompt it created.
00:08:23It's using GPT-5 and sends the newsletter,
00:08:27so let's see if it works on the first shot.
00:08:30Hopefully it doesn't.
00:08:31So we can kind of show off some of the troubleshooting back and forth.
00:08:35OK, so right here we got an error.
00:08:38What is the error?
00:08:41Bad request, unsupported parameter temperature. OK.
00:08:46Or bring this up, OK, so we shouldn't be passing temperature,
00:08:51but let's say I didn't even know how to just stop that.
00:08:54All I would have to do.
00:08:57Let's copy the output.
00:09:00Got this error.
00:09:04Paste it in.
00:09:06All right, so it said it fixed it, so let's try rerunning it.
00:09:09Obviously overkill to put the sort of error message into cloud code
00:09:14to fix a simple thing like temperature, but just wanted to demonstrate it.
00:09:18And now we get a successful output.
00:09:21The email also sends off just fine.
00:09:24And here's our little newsletter. Super easy.
00:09:27So like you saw there, that took what, five minutes to create
00:09:31this whole thing and get it working, even with a little bit of troubleshooting.
00:09:34Not bad. So to wrap this up, I think this is a great addition
00:09:38for anybody who is still trying to implement any of them workflows
00:09:41into their coding stack.
00:09:43I don't think any of them should be the thing you're using most often,
00:09:46especially for more complicated things, just writing straight up code
00:09:49inside of cloud code or codex is going to be better.
00:09:53But there are situations where it is great.
00:09:55And up until now, there hasn't been such a streamlined solution as this one.
00:10:01And it's so easy to set up and use like you just saw.
00:10:03So as always, let me know what you thought in the comments.
00:10:06And speaking of comments in a pin comment, there's a link to chase
00:10:09a plus if you want to get your hands on my cloud code masterclass.
00:10:12But besides that, I'll see you around.