n8n-nodes-mcp-session/README.md

MCP nodes for n8n with session management support

[![Verified on MseeP](https://mseep.ai/badge.svg)](https://mseep.ai/app/bd76f121-1c8f-4f5d-9c65-1eac5d81b6af)

# n8n-nodes-mcp-session

> **Important Note:**
> The Server-Sent Events (SSE) transport is deprecated and replaced by the new HTTP Streamable transport. SSE remains available for legacy compatibility, but HTTP Streamable is now the recommended method for all new implementations.

This is an **enhanced fork** of the original n8n-nodes-mcp-client that adds **advanced session management capabilities** for stateful MCP server interactions.

## 🚀 Key Differences from Original Version

### **Session Management (NEW)**
Unlike the original version that creates a new connection for every tool call, this enhanced version includes:

- **Connection Reuse**: Automatically reuses connections across multiple tool calls within the same session
- **State Persistence**: Maintains server state between operations (perfect for playwright-mcp browser automation)
- **Smart Session Management**: Sessions are automatically managed with configurable timeouts (30 minutes default)
- **Performance Optimization**: Significantly reduces connection overhead and improves response times
- **Memory Efficiency**: Automatic cleanup of expired sessions to prevent memory leaks

### **Enhanced Playwright Support**
Specifically optimized for playwright-mcp and similar stateful servers:

- **Browser State Persistence**: Keep browser instances alive between tool calls
- **Multi-step Automation**: Execute complex browser workflows across multiple nodes
- **Session Isolation**: Different workflows use separate browser sessions
- **Automatic Cleanup**: Browser instances are cleaned up after session timeout

### **Backward Compatibility**
- **100% Compatible**: All existing workflows continue to work without modification
- **No Breaking Changes**: Existing configurations and credentials remain valid
- **Seamless Migration**: Simply replace the original node with this enhanced version

### **Technical Improvements**
- **McpSessionManager**: New session management system with connection pooling
- **Error Resilience**: Better error handling and connection recovery
- **Resource Management**: Intelligent cleanup of connections and resources
- **Performance Monitoring**: Built-in session statistics and monitoring

---

MCP is a protocol that enables AI models to interact with external tools and data sources in a standardized way. This node allows you to connect to MCP servers, access resources, execute tools, and use prompts. With the enhanced session management feature, connections are automatically reused across multiple tool calls, making it perfect for stateful MCP servers like playwright-mcp that need to maintain browser state between operations.

[n8n](https://n8n.io/) is a [fair-code licensed](https://docs.n8n.io/reference/license/) workflow automation platform.

[Installation](#installation)
[Credentials](#credentials)
[Environment Variables](#environment-variables)
[Operations](#operations)
[Using as a Tool](#using-as-a-tool)
[Compatibility](#compatibility)
[Resources](#resources)

## Getting Started

Official Quickstart Video:

[![MCP Client Node Quickstart](/assets/mcp-n8n.png)](https://youtu.be/1t8DQL-jUJk)



### Community Videos

Shoutout to all the creators of the following n8n community videos that are great resources for learning how to use this node:

- [Is MCP the Future of N8N AI Agents? (Fully Tested!)](https://youtu.be/sb5hCcFYPIE)
- [Connect N8N AI Agents to EVERYTHING using MCP?](https://youtu.be/tTDRgkD-120)
- [Build an AI Agent That Can Use Any Tool (MCP in n8n Tutorial)](https://www.youtube.com/watch?v=SVZe2rdhYmA)
- [The NEW N8N MCP is an Absolute Game-Changer (Brave Search MCP)](https://youtu.be/RxXS_FpJyGM)
- [MCP & n8n Automation: The Ultimate Guide for MCP AI Agents (2025)](https://www.youtube.com/watch?v=mbQsnrxHPwE)
- [REVOLUÇÃO na criação de AGENTES no N8N com o MCP Server!!!](https://www.youtube.com/watch?v=zgH85dJcs5c) (Portuguese)

If you have a great video that you'd like to share, please let me know and I'll add it to the list!

#### Interested a deeper dive into MCP?

Check out my YouTube Series [MCP Explained](https://www.youtube.com/playlist?list=PLjOCx_PNfJ4S_oOSqrMi6t9_x1GllvQZO) for more information about the Model Context Protocol.

### Security Assessment
[![MseeP.ai Security Assessment Badge](https://mseep.net/pr/nerding-io-n8n-nodes-mcp-badge.png)](https://mseep.ai/app/nerding-io-n8n-nodes-mcp)


## Installation

### Installing from npm

To install this enhanced MCP client with session management support, use the following package name:

```bash
npm install n8n-nodes-mcp-session
```

### Installing in n8n Community Nodes

1. **Via n8n UI (Recommended)**:
   - Go to **Settings** → **Community Nodes**
   - Click **Install a community node**
   - Enter the package name: `n8n-nodes-mcp-session`
   - Click **Install**

2. **Via Environment Variable**:
   ```bash
   export N8N_NODES_INCLUDE="n8n-nodes-mcp-session"
   ```

3. **Via Docker**:
   ```yaml
   environment:
     - N8N_NODES_INCLUDE=n8n-nodes-mcp-session
   ```

### Important Notes

- This package replaces the original `n8n-nodes-mcp` with enhanced session management
- All existing workflows remain compatible - no migration needed
- Follow the [installation guide](https://docs.n8n.io/integrations/community-nodes/installation/) in the n8n community nodes documentation for detailed instructions

Also pay attention to Environment Variables for [using tools in AI Agents](#using-as-a-tool). It's mandatory to set the `N8N_COMMUNITY_PACKAGES_ALLOW_TOOL_USAGE` environment variable to `true` if you want to use the MCP Client node as a tool in AI Agents.

## Credentials

The MCP Client node supports three types of credentials to connect to an MCP server:

### Command-line Based Transport (STDIO)

![MCP Client STDIO Credentials](./assets/credentials.png)

- **Command**: The command to start the MCP server
- **Arguments**: Optional arguments to pass to the server command
- **Environment Variables**: Variables to pass to the server in NAME=VALUE format

### HTTP Streamable Transport (Recommended)

- **HTTP Streamable URL**: The HTTP endpoint that supports streaming responses (e.g., http://localhost:3001/stream)
- **Additional Headers**: Optional headers to send with requests (format: name:value, one per line)

HTTP Streamable is the recommended and modern method for all new integrations, providing better efficiency and flexibility compared to SSE.

#### Example: Using a Local MCP Server with HTTP Streamable

This example shows how to connect to a locally running MCP server using HTTP Streamable:

1. Start a local MCP server that supports HTTP Streamable:
   ```bash
   npx @modelcontextprotocol/server-example-streamable
   ```

2. Configure MCP Client credentials:
   - In the node settings, select **Connection Type**: `HTTP Streamable`
   - Create new credentials of type **MCP Client (HTTP Streamable) API**
   - Set **HTTP Streamable URL**: `http://localhost:3001/stream`
   - Add any required headers for authentication

3. Create a workflow using the MCP Client node:
   - Add an MCP Client node
   - Set the Connection Type to `HTTP Streamable`
   - Select your HTTP Streamable credentials
   - Execute the workflow to see the results

### Server-Sent Events (SSE) Transport (Deprecated, still available for legacy use)

- **SSE URL**: The URL of the SSE endpoint (default: http://localhost:3001/sse)
- **Messages Post Endpoint**: Optional custom endpoint for posting messages if different from the SSE URL
- **Additional Headers**: Optional headers to send with requests (format: name:value, one per line)

> **Deprecated:** SSE is deprecated and will not receive further updates, but remains available for legacy compatibility. For new projects, use HTTP Streamable.

#### Example: Using a Local MCP Server with SSE (legacy)

This example shows how to connect to a locally running MCP server using Server-Sent Events (SSE):

1. Start a local MCP server that supports SSE:
   ```bash
   npx @modelcontextprotocol/server-example-sse
   ```

2. Configure MCP Client credentials:
   - In the node settings, select **Connection Type**: `Server-Sent Events (SSE)`
   - Create new credentials of type **MCP Client (SSE) API**
   - Set **SSE URL**: `http://localhost:3001/sse`
   - Add any required headers for authentication

3. Create a workflow using the MCP Client node:
   - Add an MCP Client node
   - Set the Connection Type to `Server-Sent Events (SSE)`
   - Select your SSE credentials
   - Execute the workflow to see the results

> **Note:** For new projects, HTTP Streamable is strongly recommended.

## Environment Variables

The MCP Client node supports passing environment variables to MCP servers using the command-line based transport in two ways:

### 1. Using the Credentials UI

You can add environment variables directly in the credentials configuration:

![Environment Variables in Credentials](./assets/credentials-envs.png)

This method is useful for individual setups and testing. The values are stored securely as credentials in n8n.

### 2. Using Docker Environment Variables

For Docker deployments, you can pass environment variables directly to your MCP servers by prefixing them with `MCP_`:

```yaml
version: '3'

services:
  n8n:
    image: n8nio/n8n
    environment:
      - MCP_BRAVE_API_KEY=your-api-key-here
      - MCP_OPENAI_API_KEY=your-openai-key-here
      - MCP_CUSTOM_SETTING=some-value
    # other configuration...
```

These environment variables will be automatically passed to your MCP servers when they are executed.

### Example: Using Brave Search MCP Server

This example shows how to set up and use the Brave Search MCP server:

1. Install the Brave Search MCP server:
   ```bash
   npm install -g @modelcontextprotocol/server-brave-search
   ```

2. Configure MCP Client credentials:
   - **Command**: `npx`
   - **Arguments**: `-y @modelcontextprotocol/server-brave-search`
   - **Environment Variables**: `BRAVE_API_KEY=your-api-key` Add a variables (space comma or newline separated)

3. Create a workflow that uses the MCP Client node:
   - Add an MCP Client node
   - Select the "List Tools" operation to see available search tools
   - Add another MCP Client node
   - Select the "Execute Tool" operation
   - Choose the "brave_search" tool
   - Set Parameters to: `{"query": "latest AI news"}`

![Brave Search Example](./assets/brave-search-example.png)

The node will execute the search and return the results in the output.

### Example: Multi-Server Setup with AI Agent

This example demonstrates how to set up multiple MCP servers in a production environment and use them with an AI agent:

1. Configure your docker-compose.yml file:

```yaml
version: '3'

services:
  n8n:
    image: n8nio/n8n
    environment:
      # MCP server environment variables
      - MCP_BRAVE_API_KEY=your-brave-api-key
      - MCP_OPENAI_API_KEY=your-openai-key
      - MCP_SERPER_API_KEY=your-serper-key
      - MCP_WEATHER_API_KEY=your-weather-api-key

      # Enable community nodes as tools
      - N8N_COMMUNITY_PACKAGES_ALLOW_TOOL_USAGE=true
    ports:
      - "5678:5678"
    volumes:
      - ~/.n8n:/home/node/.n8n
```

2. Create multiple MCP Client credentials in n8n:

   **Brave Search Credentials**:
   - Command: `npx`
   - Arguments: `-y @modelcontextprotocol/server-brave-search`

   **OpenAI Tools Credentials**:
   - Command: `npx`
   - Arguments: `-y @modelcontextprotocol/server-openai`

   **Web Search Credentials**:
   - Command: `npx`
   - Arguments: `-y @modelcontextprotocol/server-serper`

   **Weather API Credentials**:
   - Command: `npx`
   - Arguments: `-y @modelcontextprotocol/server-weather`

3. Create an AI Agent workflow:
   - Add an AI Agent node
   - Enable MCP Client as a tool
   - Configure different MCP Client nodes with different credentials
   - Create a prompt that uses multiple data sources

![Multi-Server Setup](./assets/multi-server-example.png)

Example AI Agent prompt:
```
I need you to help me plan a trip. First, search for popular destinations in {destination_country}.
Then, check the current weather in the top 3 cities.
Finally, find some recent news about travel restrictions for these places.
```

With this setup, the AI agent can use multiple MCP tools across different servers, all using environment variables configured in your Docker deployment.

## Operations

The MCP Client node supports the following operations:

![MCP Client Operations](./assets/operations.png)

- **Execute Tool** - Execute a specific tool with parameters
- **Get Prompt** - Get a specific prompt template
- **List Prompts** - Get a list of available prompts
- **List Resources** - Get a list of available resources from the MCP server
- **List Tools** - Get a list of available tools
- **Read Resource** - Read a specific resource by URI

### Example: List Tools Operation

![List Tools Example](./assets/list-tools.png)

The List Tools operation returns all available tools from the MCP server, including their names, descriptions, and parameter schemas.

### Example: Execute Tool Operation

![Execute Tool Example](./assets/execute-tool.png)

The Execute Tool operation allows you to execute a specific tool with parameters. Make sure to select the tool you want to execute from the dropdown menu.

## Using as a Tool

This node can be used as a tool in n8n AI Agents. To enable community nodes as tools, you need to set the `N8N_COMMUNITY_PACKAGES_ALLOW_TOOL_USAGE` environment variable to `true`.

### Setting the Environment Variable

**If you're using a bash/zsh shell:**
```bash
export N8N_COMMUNITY_PACKAGES_ALLOW_TOOL_USAGE=true
n8n start
```

**If you're using Docker:**
Add to your docker-compose.yml file:
```yaml
environment:
  - N8N_COMMUNITY_PACKAGES_ALLOW_TOOL_USAGE=true
```

**If you're using the desktop app:**
Create a `.env` file in the n8n directory:
```
N8N_COMMUNITY_PACKAGES_ALLOW_TOOL_USAGE=true
```

**If you want to set it permanently on Mac/Linux:**
Add to your `~/.zshrc` or `~/.bash_profile`:
```bash
export N8N_COMMUNITY_PACKAGES_ALLOW_TOOL_USAGE=true
```

Example of an AI Agent workflow results:

![AI Agent Example](./assets/execute-tool-result.png)

After setting this environment variable and restarting n8n, your MCP Client node will be available as a tool in AI Agent nodes.

## Compatibility

- Requires n8n version 1.0.0 or later
- Compatible with MCP Protocol version 1.0.0 or later
- Supports both STDIO and SSE transports for connecting to MCP servers
- SSE transport requires a server that implements the MCP Server-Sent Events specification

## Resources

* [n8n community nodes documentation](https://docs.n8n.io/integrations/community-nodes/)
* [Model Context Protocol Documentation](https://modelcontextprotocol.io/docs/)
* [MCP TypeScript SDK](https://github.com/modelcontextprotocol/typescript-sdk)
* [MCP Transports Overview](https://modelcontextprotocol.io/docs/concepts/transports)
* [Using SSE in MCP](https://github.com/modelcontextprotocol/typescript-sdk/blob/main/src/client/sse.ts)