n8n MCP: what it is and how the two fit together
By the Wobble team
n8n MCP means using the Model Context Protocol inside n8n, and n8n supports it in both directions. A workflow can be published as an MCP server, so an AI assistant can call it as a tool. An n8n AI agent can also act as an MCP client and call tools that live on someone else's MCP server. MCP is the shared connector. n8n is where the actual steps run.
Does n8n use MCP?
Yes, through two dedicated nodes, one for each direction.
| Direction | Node | What happens | Use it when |
|---|---|---|---|
| n8n as MCP server | MCP Server Trigger | Your workflow is offered to an AI client as a named tool with defined inputs. | You already have a working process, such as looking up an order or logging a lead, and want an assistant to trigger it on request. |
| n8n as MCP client | MCP Client Tool | An agent inside n8n reaches out to an MCP server and uses the tools it lists. | The tool you need already exists as an MCP server and you would rather not rebuild the connection by hand. |
Why a protocol instead of a custom connection
Before MCP, every pairing of an AI model and a tool needed its own glue. A tool built for one assistant did not work with another. MCP standardises how a tool describes itself and how a model calls it, so one server can be used by many clients. That is the practical gain: you describe the tool once.
It also changes who does the deciding. In a normal n8n workflow, you fix the order of steps in advance. With MCP, the model picks which tool to call and when. That is useful for open-ended requests and a poor fit for anything that must happen the same way every time.
Working examples
- Expose a workflow as a tool called check_stock. It takes one input, a product code, and returns a plain answer such as "In stock, ready to ship" or "Out of stock, next delivery due Friday". An assistant in a chat window calls it and relays that answer.
- Point an n8n agent at a GitHub MCP server so it can read issues or open a pull request as part of a larger flow.
- Run the agent's reasoning on an OpenAI model while the tools it calls sit behind MCP servers you control. The model and the tool layer are separate choices.
- Host your own MCP server alongside the workflows, so the server and the logic that backs it stay in one place.
What can go wrong
An MCP server is a door into your systems, and three decisions need making before it opens: who can reach the server, whether it requires a credential, and which tools are exposed at all. Anything you expose can be called by whatever client connects, and a model can call it in ways you did not picture. Keep each tool narrow, give it only the permissions it needs, and put approval in front of anything that writes, sends or spends.
Tool descriptions matter more than people expect. The model chooses tools by reading them, so a vague description leads to wrong calls. Test with real requests, read what the agent actually did, and expect to rewrite descriptions more than once.
MCP also does not remove the need for a clean process underneath. If the workflow behind a tool is unreliable, wrapping it in a protocol only makes the unreliability easier to reach. Approval steps on finance workflows are one place this shows up, as in accounting automation in Pakistan.
How we would run an n8n MCP build
We would design the workflow first, with fixed steps and clear inputs. Then we would log every run, check the results against what the process should do, and only then add the AI layer, so the model sits on top of something that already behaves. Where several agents share tools, AI agent orchestration covers how they are coordinated.
Common questions
Do I need to write code to use n8n MCP?
Not to start. The nodes are configured in the n8n editor. Code becomes useful when you write your own MCP server or need custom handling of inputs.
Is n8n MCP the same as an n8n AI agent?
No. An agent is the reasoning step. MCP is one way that agent reaches tools. You can use an agent without MCP, and you can expose workflows over MCP without building an agent in n8n.
Does MCP change how assistants find my business?
MCP lets an assistant act through your tools, which is different from being named in its answers. For the second question, see AI answer engine visibility for brands.