Loading video player...
NN just released instance level MCPs and
in this video I'm going to walk you
through why you should care and why more
people should be talking about it. This
new layer of MCP not only lets you
execute and audit your existing
workflows in NIDN, but you can also edit
them in a way that's a lot cleaner and
buttery compared to the open-source
alternatives. I'll show you how we can
enable it in NADN and how you can use it
in things like cloud desktop and even
cursor and cloud code. Let's dive right
in. All right. So when you pop into my
NADN, you can see that these two
workflows here are not only toggled to
on, but we have two different MCP icons
next to the workflow. And this shows you
that the MCP server has access to
execute and edit both of these
workflows. To enable this, you just have
to do two things. Thing numero uno is
you have to go to your admin panel and
make sure that you're updated on the
latest version. I'm running on the
latest beta version and that is the
1.122.3.
And then once you update it, you should
be able to see a new menu. And when you
navigate back and go to the left hand
side right here and click on this icon,
go to settings and then pop into this
new tab called MCP access. Yours should
be completely blank and then it should
have a server URL with the server
associated with your edit end URL. Now,
you can see right here I already have my
two available workflows that we saw on
the front end. And all you have to do is
click on this access token tab. And then
you'll see right here it shows your
access token. You can click on this
little dropdown and then it'll reveal
the general structure of the MCP server
connection. All you have to do to
configure yours is click on this refresh
button and then this will populate right
here with your key. Then you can just
copy this and hold this on the side
because we'll need to use this in cloud
desktop in cursor in cloud code or
wherever you want to be able to invoke
the functions of this MCP server. And
one last thing I forgot to mention is to
make sure you enable the MCP. Otherwise,
you won't be able to do this to begin
with. Once you have this configured,
then you can go back to your Niten
homepage and click on any workflow you
want and then set it to active to
include it in the MCP server. And then
you'll be able to click on this enable
MCP access. Once you click this, you'll
see this icon pop up right here. And
that means that you can now access this
via that MCP layer. Now, if you've ever
worked with MCPS, you know that they
like metadata or context. So, what you
can do for any workflow that you enable
is click on this icon right here. And
this edits the workflows description. It
doesn't do anything functional for the
workflow, but you basically add a bunch
of labels or information about what this
workflow does. So if you try to invoke
it from claw desktop from cursor
wherever it will have a higher
likelihood to find the right workflow
and especially not get confused between
very similar functioning workflows. So
in this case this workflow is for the
Gemini rag API. So I could say something
like this workflow is designed to help
create a rag implementation workflow
using the brand new Gemini rag API. So
in this case then we have this we can
click save. And now if we go back to our
original screen there and go to settings
and go to MCP access, you can see right
here this is the workflow and here's the
description we just added right now. Now
what does this look like to invoke from
something like cloud desktop? Well, if
you remember before we had this JSON URL
you copy to your clipboard. All we have
to do if we go into something like
Claude is go into the settings and then
click on developer right here. And then
this is where you can configure your
JSON file for enabling the MCP server.
Once you click on edit config, you want
to paste in the exact URL you got from
NIDEN. And when you open up the file,
you should have something like this
where you visibly have the NADN MC file
right here. And this is where your API
key will go. All you have to do again is
just copy paste as is. In case you had
other MCV servers, what I would do is
take your existing file and then put it
into something like claude with the new
URL you've copied over from edit and
then tell it make one compound MCP file
that I can easily copy paste. Once you
put that together, then save the file,
exit completely and exit claw desktop
and when you log in, you should see the
brand new MCP server. So now if I go on
my search and tools section right here,
you should see I now have this ended in
MCP and these are the three tools like I
mentioned. You can search workflows, you
can execute workflows, and you can get
workflow details where you can also
submit changes to those same workflows
as well. So let's say I wanted to
simulate running this workflow. If we go
back into cloud code, all I have to say
is execute the endmc with a dummy
question to ping the workflow that has a
web hook chatbot. Now, in this case, I
just described briefly what it looks
like. It's then able to search the
workflows. And you'll see once I send
that query, it invokes this search
workflows function. It sends this query
right here that says web hook chat. And
then at the bottom, you can see the
response of the actual workflow itself.
And you can see the trigger, the lang
chain component, which is the AI agent.
It then searches it, gets the workflow
details, so more information about that
workflow, and then actually executes the
request itself. So this is executing the
hello. This is a ping test. What is 2
plus2? Comes back with a response at the
very bottom. You can see right here you
get a success. And then it responds with
the success in plain language right
here. And now if I say send a more
philosophical question instead of having
to now retrieve the workflow again since
it's in context, we can take advantage
of that and it just re-executes the same
workflow because it knows exactly where
it's going. So this one sends if a tree
falls in a forest and no one is around
to hear it doesn't make a sound then
comes back with a response and this is
what it comes back with.
Now this is cloud desktop. I'll show you
how you can use this in something like
cursor cloud code where you can ask it
to also edit and fix the workflow if
it's not functioning. So here we are in
cursor and you'll see right here I have
my MCP JSON where I've pasted the exact
same JSON I just showed you before. And
then if you want to use it in cursor
then we can click on and write MCP.
We'll click on tools and MCPS. You'll
see right here I've enabled the MCP
server from Naden. And I also have the
OG NIN MCP server. The one that is open
source. It helps you build workflows
without having to pay for anything. Now
I did compare both if you're curious. If
you use the open-source NAN MCP, you can
totally do tons of damage. The one thing
I noticed though is on editing, the
editing is a lot more tailored, stable,
and robust if you use the actual new
native MCP server. So now that this is
set up in cursor, let's simulate
building a very basic workflow which is
purposely incomplete and let's have the
niten MCP complete it until it can
finally execute it. I'm going to build
this exact super basic workflow from
scratch. So if I go into plus, I go into
workflow and let's create a brand new
one. Let's just do a build with AI
because I'm lazy. Can you create a very
basic workflow where you have an input
web hook that's expecting some form of
query or a question and then it goes to
an AI agent that has let's say GPT4.1
mini running behind the scenes to answer
the question. Then we have a respond to
web hook and we also have the window
buffer memory so it can keep the track
of the conversation. So other than the
broken English there, it should be able
to understand that and put it together.
And once it does that, I'm going to tell
you straight up where it would be
missing information from being able to
run. And I'll keep it like that on
purpose to show you the power of this
MCP server. So this is the workflow that
we saw before and I'll tell you where
all the flaws are. So we set up a
duplicate of what you saw before, but
this one is missing certain pieces.
Number one, if you have a web hook, that
web hook is going to have a series of
information. the names of the fields
passing through will be something we
have to map in the AI agent field and
the window buffer memory will have to
know to get the information from the AI
agent and that there's something called
a session ID. All of this right now is
missing. It's completely blank. All we
have I assume is some form of JSON query
right here. And then even the respond to
web hook won't work. So, if I just add a
description and let's say this is the
brand new chatbot that is incomplete
that we have to complete and make sure
it works properly. Okay. So, if we add
this and one thing I'll do to make my
life easier is copy the exact name just
so I can make the MCP work a little bit
less. So, once we go back into cursor
here, I'll say I want to be able to go
and execute the following workflow.
Chances are it's not properly
configured. It might be missing some
pieces from being able to run this
chatbot so we can ping over a web hook
or a web hook input and then send that
over to the agent to come back with a
response. Here's the exact name of the
workflow. I need you to use the naden
mcp http server that I've attached
behind the scenes. So this should try to
execute the execute workflow function.
It won't work guaranteed because it's
missing pieces. But once it realizes it
doesn't work, and I gave it a little bit
of a preview that I don't expect it to
work, it should go and iterate on its
own until it maps the fields correctly
until it can execute it. So you can see
here, it's first going to query the
exact workflow name. It should find it
since it's the exact verbatim title.
There you go. You should see it found it
right here. I have this set to YOLO
mode, so it'll keep running on its own,
but it'll ask me for permission on the
big things. So it's now going to plan
what it looks like. It sees the web
hook, the AI agent, the respond to web
hook. It should realize it's not going
to be able to execute and basically add
all the different pieces we need to have
it run. So you can see right here, it's
already identified two issues. One, our
web hook is set to get, not post. And
number two, the web hook response mode
is immediate. It needs to change it to
make it work from end to end. So click
on run here, and I'll update you after
it's ran through, let's say, 30 to 40%
of the way there. So here are the
parameters it's going to send to adjust
the API request. Then we click on run.
It should now start to literally
configure the JSON behind the scenes to
edit the workflow. So if I keep clicking
on run here, you can literally see if I
do show more, this is the underlying
JSON for the workflow, which is
something I used to use manually with
cloud or cloud projects back in the day,
but now it's native. And what's
different about this versus the OG in an
MCP is it's a lot more tailored to our
exact workflow. and our preferences that
we have for the types of nodes that
we're using. And you can see right here,
it comes up with yet another issue that
I told you would be a problem, which is
a missing session ID from the memory of
our chatbot. So, when we click on run,
it should now keep adding more
parameters until we get to the final
part where fingers crossed we should be
able to run it.
And in under a minute, you could see we
finally got the workflow is now working.
So, it fixed all the missing IDs. It
fixed any prompting issues. it logged in
or used my credentials I already had in
my account for OpenAI. And while I'm
showing you the foundations in this
video, you can use this MCP server for
all kinds of things. Just very important
things to remember is if you want to use
any workflow, make sure it's toggled to
on, you enable it via the MCP, and most
importantly, you want to be very sure
about whatever changes you're making,
you're comfortable with pushing that to
production if it's a productionbased
workflow. If you found this video
helpful, please let me know down in the
comments below. helps the video, helps
the channel, and I'll see you in the
next
Join My Community to Level Up ➡ https://www.skool.com/earlyaidopters/about 📅 Book a Meeting with Our Team: https://bit.ly/3Ml5AKW 🌐 Visit Our Website: https://bit.ly/4cD9jhG 🎬 Core Video Description n8n just quietly shipped instance-level MCP—and it’s a massive unlock if you’re building serious automation or AI agents. In this ~11-minute walkthrough, I’ll show you exactly how to turn your n8n instance into a first-class MCP server so you can search, execute, and even edit workflows directly from tools like Claude Desktop, Cursor, and Claude Code. You’ll see why this native MCP layer is smoother, more stable, and way more “buttery” to work with than the open-source alternatives—especially when it comes to live-editing real production workflows. We’ll start by enabling MCP Access inside n8n, grabbing your instance-level MCP config and access token, and wiring it into Claude Desktop via a single JSON file. From there, I’ll demo how Claude can automatically discover your workflows, inspect their structure, and execute them end-to-end—complete with webhook triggers, LangChain/AI agent nodes, and responses flowing back in plain language. Then we’ll jump into Cursor, where I compare the original open-source n8n MCP with the new native instance-level MCP and show why the new one is dramatically better at editing workflows. To prove it, I intentionally build a broken chatbot workflow: a webhook → GPT-4.1 mini agent → memory → webhook response chain that’s missing critical pieces like the correct HTTP method, response mode, field mappings, and session ID wiring. Using only the n8n MCP server, I ask Cursor/Claude to execute it, watch it fail, and then let MCP iteratively patch the underlying JSON until the workflow fully runs—fixing IDs, modes, memory wiring, and credentials along the way. By the end, you’ll know how to: Turn your n8n instance into an MCP server with instance-level access Expose specific workflows (with metadata and descriptions) so AI tools reliably pick the right one Plug n8n MCP into Claude Desktop and Cursor using a simple JSON config Search, execute, and auto-debug workflows purely via MCP tools Safely use MCP-driven editing on production workflows while staying in control of what goes live If you’re using n8n as the backbone of your AI systems, this is how you connect it directly to your AI coding environment and assistants—so they can not only call your workflows, but repair and evolve them with you. ⏳ TIMESTAMPS: 00:00 – Intro: Why instance-level MCP in n8n actually matters 00:28 – Visual tour: MCP-enabled workflows and new icons in n8n 00:47 – Updating n8n & finding the new MCP Access settings tab 01:21 – Configuring MCP access: server URL, access token, and JSON 02:01 – Enabling workflows for MCP & adding rich descriptions/metadata 02:59 – Wiring n8n MCP into Claude Desktop via the JSON config file 03:43 – Demo: Searching and executing workflows from Claude using MCP 05:45 – Using the n8n MCP server inside Cursor (vs the open-source MCP) 06:36 – Building an intentionally broken chatbot workflow in n8n 07:59 – Asking MCP (via Cursor) to run the broken workflow and iterate 09:00 – Auto-fixing webhook mode, response behavior, and session ID 10:09 – Final run: working chatbot workflow and how MCP edited the JSON 10:47 – Best practices, production cautions, and closing thoughts #n8n #n8nmcp #MCP #ModelContextProtocol #ClaudeDesktop #CursorIDE #AIWorkflows #WorkflowAutomation #AIEngineering #DevTools #instancelevelmcp #ChatbotDevelopment #OpenAI #GPT41 #ClaudeAI #AutomationTools #nocode #claudecode