n8n MCP: Server Trigger, Client Tool, and Instance MCP
n8n does MCP in three different directions, and picking the wrong one is the usual reason a setup goes nowhere. It can act as an MCP server exposing your workflows to AI clients. It can act as an MCP client consuming someone else’s server. And separately from both, it ships an instance-level MCP server that lets an AI client build workflows inside n8n.
Three features, three purposes, similar names. Here is which is which.
MCP Server Trigger: your workflows become tools
The MCP Server Trigger node turns n8n into an MCP server. It is an entry point rather than an action: it exposes a URL that MCP clients connect to, and the tools attached to it become callable by whatever assistant you point at that URL.
The node shows two URLs at the top of its panel, test and production. Using the test URL in a client and then wondering why nothing works after you close the editor is a common first mistake. Test URLs are live only while you are listening in the canvas.
This is the direction you want when the goal is “let Claude or ChatGPT run the automation I already built in n8n.”
MCP Client Tool: external servers become tools for your agent
The MCP Client Tool node is the other direction. It consumes tools from an external MCP server so an n8n AI Agent can call them mid-workflow.
Note the naming trap: there is both an MCP Client node and an MCP Client Tool node. If you are attaching tools to an AI Agent, the Tool variant is the one you want. If the service you are after already appears in n8n’s MCP servers registry, you can connect it straight from the node panel without setting up a credential. The Client Tool node is for servers that are not in the registry yet.
Instance-level MCP: build workflows from your AI client
Separate from both nodes, n8n ships a native instance-level MCP server, in public preview since April 2026 and available on Cloud, Enterprise, and self-hosted Community Edition. Point Claude Desktop, ChatGPT, Cursor, or any MCP client at it and the assistant can build, test, and publish workflows directly inside your instance.
This one confuses people most, because it looks like the Server Trigger and does something quite different. The Server Trigger exposes a workflow you built. The instance server exposes n8n itself as a thing to be operated.
Where MCP stops, in n8n and everywhere
There is no such thing as an MCP trigger. The protocol's own working group says plainly that clients learn about server-side changes by polling for them, and a real event mechanism remains unshipped. Carly is where that part lives, watching thousands of apps and starting the work itself.
n8n’s own triggers are a separate mechanism and they work well. The point is narrower: the MCP layer connecting your assistant to n8n is request and response. It carries a call when the assistant makes one. It does not carry an event back the other way on its own.
When you would rather not maintain the plumbing
n8n rewards people who enjoy building the graph. If what you actually want is the outcome, Carly is the same job from the other end: you describe the rule in plain English, it interviews you, and it builds the workflow with you.
- Add your accounts to Carly. Sign in at carlyassistant.com, open Integrations, find the service, click Connect, and authorize. Click Connect again for the next one and keep going for as many as you want. Carly connects to thousands of apps, so mail, calendars, files, and your CRM sit behind one connector.
- Add Carly to the assistant you already use. Carly publishes an MCP server at
https://carlyassistant.com/mcp/. In Claude it goes under Customize → Connectors. In ChatGPT it goes in Plugins. No node graph either way. - Let it run on the event. Workflows start when the message lands rather than when you open something.
Free Zapier-style workflows; AI agents from $35/month, and no server to host.
Frequently Asked Questions
What is the difference between the MCP Server Trigger and MCP Client Tool nodes?
The Server Trigger makes your n8n workflows available to outside MCP clients. The Client Tool lets an n8n AI Agent call tools from an outside MCP server. They point in opposite directions.
Why does my n8n MCP Server Trigger stop working?
Most often because the client is pointed at the test URL rather than the production one. The node displays both, and the test URL only listens while the canvas is open.
Should I use the MCP Client node or the MCP Client Tool node?
Use MCP Client Tool when you are giving tools to an AI Agent. That is what it exists for.
Can an AI assistant build n8n workflows for me?
Yes, through the instance-level MCP server in public preview, which lets clients like Claude Desktop, ChatGPT, or Cursor build, test, and publish workflows in your instance.
Do I need to self-host anything to connect an assistant to my tools?
Not necessarily. A hosted MCP server such as Carly is a URL you paste into your assistant, with no instance to run or update.
More: Claude MCP · ChatGPT MCP · n8n alternatives · MCP servers: the complete list · What is MCP?
Ready to automate your busywork?
Carly schedules, researches, and briefs you—so you can focus on what matters.
See what people say
"Before Carly, I relied on a Calendly link, but the whole process felt impersonal and not very professional. Carly changed that by handling all the back-and-forth, so I'm no longer stuck in endless email threads trying to line up schedules.
Now Carly reaches out to candidates, shares my real-time availability, lets them pick a slot, then sends a Zoom link and drops it straight into my calendar. She sends reminders to both of us before each call, which has significantly reduced no-shows and last-minute confusion.
On top of scheduling, Carly acts like a full executive assistant, sending me my schedule the night before so I can prepare for each call. It reminds me of the old x.ai assistant, but Carly is noticeably smarter, faster, and better suited to my healthcare recruitment business."
