Resources · n8n integration

n8n MCP Client Tool nodeConfigure MCP Client Tool in n8n.

An MCP server already holds tools your agent could be calling. The n8n MCP Client Tool node connects to that server over an SSE endpoint and hands its tools straight to an AI Agent. Four parameters, one output port and no main input: it is a sub-node, and it only acts when the agent above it asks.

Try the MCP Client Tool node on n8nSee the 4 parameters

Verified Trustpilot reviews · AI, automation & growth agency

Why automate

What does the n8n MCP Client Tool node give an agent?

The node is a Model Context Protocol client. It talks to an external MCP server, reads the tools that server publishes, and exposes them to the AI Agent it is wired to. Your agent stops being limited to the tools you built by hand in n8n and starts calling whatever that server offers, under the filter you set.

First scenario: a personal assistant. An agent runs on a chat model, keeps a short history through a memory sub-node, and reaches a calendar and a mailbox through one MCP server. That is the shape of the published template 3905, which pairs this node with a Google Gemini Chat Model and Simple Memory. If you would rather call those services directly, the Gmail node stays the shorter path.

Second scenario: an internal server that fronts your own logic. Template 3514 builds an MCP server out of HTTP Request and Call n8n Workflow Tool, then points this node at it. The agent sees a clean list of tools instead of a pile of raw endpoints, and you keep the implementation in n8n.

Third scenario: narrowing an oversized server. Some servers publish far more tools than one agent should see. Tools to Include set to selected or except cuts the list down before it ever reaches the model, which keeps the agent focused and its prompts shorter.

When to prefer something else: a dedicated node beats an MCP round trip whenever one exists. For a spreadsheet, Google Sheets is direct. For a chat post, Slack is direct. The n8n docs also point out that services already listed in the n8n MCP servers registry connect straight from the node panel, without a credential; this node is for servers that are not in the registry yet.

Known limits. The node has no main input and never runs on its own: it only acts when the root node it feeds decides to call one of the tools. It connects over an SSE endpoint, so the server has to expose one. And the AI Agent above it still needs its own chat model, from OpenAI or another provider, whose API calls that provider bills on its own terms.

Connect

How do you authenticate against your MCP server?

  1. 01

    Paste the SSE endpoint

    Open the node and fill SSE Endpoint, the SSE address of your MCP server. The field shows the expected shape as a placeholder, e.g. https://my-mcp-server.ai/sse. The value is a string, so an expression works here too when the address comes from an earlier item, for instance {{ $json.endpoint }}. Nothing else in the node works until this one is set.

  2. 02

    Pick the authentication method

    The Authentication selector offers bearerAuth (Bearer Auth), headerAuth (Header Auth) and none (None). Per the node documentation, the node also supports multiple headers and OAuth2. Bearer covers a server that expects a token; generic header covers a server with its own key name. Choose None only when the server accepts an unauthenticated connection.

  3. 03

    Create the credential once

    Credentials live in the n8n Credentials menu, not inside the node, and a credential created once is reused across every workflow on that instance. For Multiple Headers Auth, the documentation says each header goes in as a Name and Value pair, and you can add as many as the server needs, for example an API key alongside a username.

Connections

Where does this sub-node plug in?

n8n splits AI work between a root node, which receives the workflow items on its main input, and sub-nodes that clip onto its ports, one type of sub-node per port. MCP Client Tool is a sub-node: it has no main input, so it lives on the port below.

Output (what it plugs into)

  • Toolai_tool
01ToolThis is the node's only connection, and it is an output: the node attaches to the Tool port of a root node, which then decides when to call the tools the MCP server published.

This is the node's only connection, and it is an output: the node attaches to the Tool port of a root node, which then decides when to call the tools the MCP server published.

Key parameters

  • Required: a sub-node never runs alone, so an unattached MCP Client Tool does nothing at all.
  • AI Agent: the usual host, the node that reasons over a chat model and picks a tool per turn.
  • AI Agent Tool: an agent exposed as a tool to another agent, so the MCP tools travel one level deeper.
  • MCP Server Trigger: the root that republishes tools to outside clients, forwarding what this node brought in.
Use cases
wire it under an agent that already has a chat model, and the tool list from your server shows up in that agent's next run.
Parameters

Which parameters does the MCP Client Tool node expose?

The MCP Client Tool node has 4 parameters. For each one: the node as you configure it in n8n, what the parameter changes, and our field notes.

01

SSE Endpoint

sseEndpoint

What you see in n8n

Notes & use cases

Points the node at the MCP server it should read tools from. It is the one required field, and an empty required field stops the node from executing.

Key parameters

  • SSE Endpoint: a string, the SSE endpoint of your MCP server, shaped like the placeholder e.g. https://my-mcp-server.ai/sse; use {{ $json.sseEndpoint }} when the address travels with the item.
Use cases
one address per server, so a workflow that reaches two MCP servers carries two copies of this node, each with its own endpoint.
02

Authentication

authentication

What you see in n8n

Notes & use cases

Decides how the connection identifies itself to the server. Set it before your first run, because a server that expects a token simply refuses an anonymous client.

Key parameters

  • Authentication: three choices in the selector, bearerAuth (Bearer Auth), headerAuth (Header Auth) and none (None). The documentation adds multiple headers and OAuth2 as supported methods, and says None attempts the connection with no authentication at all.
Use cases
a public demo server runs on None, while an internal server usually sits behind a bearer token created in the n8n Credentials menu.
03

Tools to Include

include

What you see in n8n

Notes & use cases

Filters what the server publishes before the AI Agent ever sees it. Leave it alone on a small server, reach for it as soon as the tool list grows past what one agent should handle.

Key parameters

  • Tools to Include: the selector itself, with all (All) to expose everything, selected (Selected) to list what goes through, and except (All Except) to list what stays out.
  • Tools to Include: the multi-select that appears on selected, filled from the list or with an expression.
  • Tools to Exclude: its mirror on except; everything you do not name stays available to the agent.
Use cases
a server that both reads and writes can be cut to read-only tools for an agent that answers questions.
04

Options

options

What you see in n8n

Notes & use cases

A collection you open with Add Option, holding the settings that do not deserve a field of their own. It stays empty on most workflows.

Key parameters

  • Timeout: the time in milliseconds n8n waits for tool calls to finish. Raise it for a server that does real work behind a tool, since a slow call cut short looks exactly like a broken server.
Use cases
a tool that queries a large dataset before answering is the usual reason to open this collection at all.
Need help

Need help automating MCP Client Tool with n8n?

A person reads every message.

FAQ

Questions about the MCP Client Tool node

01Is the MCP Client Tool node included in n8n?
Yes. It ships inside n8n's AI package, @n8n/n8n-nodes-langchain, so there is nothing to install and nothing extra to pay on the n8n side. It behaves the same on n8n Cloud, the hosted offer run by n8n, and on a self-hosted instance installed through Docker or npm under the Community Edition and its Sustainable Use license. Cost comes from elsewhere: the agent above this node runs on a chat model, and the model provider bills its own API calls on its own terms.
02What do you need to make it work?
Two things. An SSE endpoint for the MCP server you want to reach, pasted into the SSE Endpoint field, and an authentication method that matches what the server expects. The Authentication selector offers Bearer Auth, Header Auth and None, and the node documentation adds multiple headers and OAuth2. Credentials are created once in the n8n Credentials menu and reused everywhere. One shortcut: a service already listed in the n8n MCP servers registry connects from the node panel without a credential.
03What are the limits of the n8n MCP Client Tool node?
It is a sub-node, version 1 in the catalog. It has no main input and never executes on its own, so it only does something when a root node calls one of its tools. It reaches a server through an SSE endpoint, which means a server without one is out of reach. Its own surface is small, four parameters, and everything else about behavior belongs to the server you connect to. Older workflows may still show an earlier version with fewer options.
04What should you connect at minimum, and where do you start?
At minimum, one root node on the Tool port. An AI Agent is the usual choice, and that agent needs a chat model of its own before anything runs, from a provider such as OpenAI, Anthropic, Google or Mistral. The published templates are a decent starting point: template 3905 pairs this node with a Google Gemini Chat Model and Simple Memory, while template 3770 adds Call n8n Workflow Tool next to it. Start with one server, All tools, and narrow later.
05n8n or Make for connecting an MCP server?
It depends on where you want the workflow to run. Make is hosted, with no self-hosting option, and bills per operation, so a chatty agent that calls many tools per turn is a cost you watch. n8n runs on Cloud or on your own machine, and a workflow is the same in both cases, which matters when the MCP server sits on an internal network your automation platform has to reach. Visual logic differs too, so try the shape you find readable.
Hack'celeration Lab

Get our weekly integration tips.

No spam. Unsubscribe anytime.

Same platform

Other docs you might need on this stack.

Browse all integrations