MCP in n8n: MCP Server Trigger and Client for Claude

n8n offers two nodes for the Model Context Protocol (MCP): the MCP Server Trigger, with which an n8n workflow itself becomes an MCP server and exposes its tools to Claude or other MCP-capable clients, and the MCP Client, with which an n8n workflow conversely uses tools from an external MCP server. Both nodes are based on the official n8n documentation on the LangChain integrations and are aimed at teams who want to make n8n automations controllable directly from an AI chat or agent, without building a dedicated interface for every application. As of: July 2026.

What is MCP, briefly explained

The Model Context Protocol is an open standard through which AI applications like Claude can access external tools, data and services, without an individual integration having to be programmed for every combination. An MCP server provides a list of tools, an MCP client, such as Claude Desktop or another MCP-capable client, queries this list and calls individual tools as needed. According to its own documentation, n8n can take on both roles: as a server that offers its own workflows as tools, and as a client that integrates tools from external MCP servers into a workflow. More about the protocol itself is explained in the official introduction to Model Context Protocol.

MCP Server Trigger: providing n8n workflows as tools

According to the n8n documentation, the MCP Server Trigger node, unlike classic triggers, does not react to individual events but instead permanently provides a URL through which MCP clients can list and call available tools. Workflows are connected to the trigger via the "Custom n8n Workflow Tool" node, which then appear as individual, nameable tools. n8n generates a test URL for the manual execution mode and, after publishing, a separate production URL, each under an automatically generated but changeable path. Bearer Auth and Header Auth are available for authentication, the same as known from the regular HTTP Request credentials in n8n.

• Transport: Server-Sent Events (SSE) and streamable HTTP are supported, not classic stdio.
• Operation in Queue Mode: With multiple replicas, according to the documentation, all requests to /mcp* must be routed to a single, dedicated webhook replica, otherwise the connections become unstable.
• Reverse Proxy: With nginx as an upstream proxy, proxy buffering must be disabled for the MCP endpoint.

For the connection to Claude Desktop, the documentation names a gateway that proxies SSE messages to a stdio-based server, for example via the command-line tool mcp-remote with the respective MCP URL and an Authorization header.

MCP Client: using external MCP servers from n8n

Conversely, the MCP Client node, according to the n8n documentation, integrates the tools of an external MCP server into a workflow. Server transport and the MCP endpoint URL of the target server are configured, the available tools are then automatically retrieved from the external server and available for selection. For authentication, the node supports Bearer token, single or multiple headers, as well as OAuth2, a connection without any authentication is also possible. In addition, it can be specified whether return values such as images or audio files are converted into binary data, and how long a tool call may take at most. Important for use with AI agents: the documentation explicitly points out that for agents the specialized MCP Client Tool node should be used, the regular MCP Client node is mainly suited for direct integration into fixed workflow steps.

Practical example

A typical scenario: an n8n workflow with MCP Server Trigger provides two custom workflow tools, one for creating a CRM contact, one for retrieving open invoices. Claude Desktop is connected as an MCP client via the trigger's production URL and can call both tools as needed in the chat, for example when an employee asks which invoices of a customer are still open. In the other direction, the same n8n workflow can use an MCP Client Tool node to connect an AI agent in n8n to tools of an external MCP server, for example to query data from a specialized system without having to develop a dedicated node for it. Anyone planning such automations with connected AI agents will find background on the topic in the article on AI Agents. Since MCP is still a comparatively young protocol, it is worth taking a look at the current n8n documentation before productive use, as details on authentication and transport may continue to evolve.

Frequently asked questions about MCP in n8n

What is the difference between MCP Server Trigger and MCP Client in n8n?

The MCP Server Trigger turns an n8n workflow into an MCP server whose tools can be called from outside, for example by Claude. The MCP Client does the opposite: n8n calls tools of an external MCP server and uses them within its own workflow.

Can I connect n8n to Claude Desktop?

Yes. According to the n8n documentation, the production URL of the MCP Server Trigger can be entered into the configuration of Claude Desktop via a gateway such as mcp-remote, so that Claude can directly call the workflow tools provided there.

What authentication does the MCP Server Trigger support?

According to the documentation, the trigger supports Bearer Auth and Header Auth, based on the same credential types also used for regular HTTP Request nodes in n8n.

Do I need a dedicated MCP node for an AI agent?

Yes. For integration into an AI agent in n8n, the documentation recommends the specialized MCP Client Tool node, not the regular MCP Client node, which is more intended for fixed workflow steps.