MCPInstall
n8n-MCP-Agent logo

n8n-MCP-Agent

by Anshumaan031 Anshumaan031/n8n-MCP-Agent

n8n workflow demonstrating an AI agent using Model Context Protocol (MCP) to dynamically discover and execute external tools from Brave Search and Convex with persistent PostgreSQL chat memory.

View on GitHub
READMEmaster

n8n MCP Agent Demo

This repository contains a workflow for n8n that demonstrates how to build an AI agent that leverages the Model Context Protocol (MCP) to access and execute external tools.

n8n MCP Agent Demo Screenshot

This is using the n8n community MCP node found here. You must download this before using this demo n8n workflow. Instructions can be found there.

Overview

The n8n MCP Agent Demo is a workflow that showcases:

  1. How to create an AI agent in n8n that can dynamically discover and use tools via MCP
  2. Integration with multiple MCP clients (Brave Search and Convex)
  3. Persistent chat memory using PostgreSQL

MCP Tool Integrations

The workflow includes two sets of MCP client tools:

  1. Brave Search Tools:

    • List Brave Tools: Discovers available Brave Search tools
    • Execute Brave Tool: Executes a specific Brave Search tool with parameters

  2. Convex Tools:

    • List Convex Tools: Discovers available Convex tools
    • Execute Convex Tool: Executes a specific Convex tool with parameters

How It Works

  1. A user message is received through the chat trigger
  2. The AI agent processes the message using the OpenAI model
  3. The agent can:
    • List available tools from Brave Search or Convex
    • Execute specific tools with parameters
    • Maintain conversation context using PostgreSQL memory

The workflow is designed with a specific system message that instructs the AI on how to:

  • Discover available tools before attempting to use them
  • Parse tool information including name, description, and parameter schema
  • Format tool parameters correctly when executing tools

System Message Details

The agent is configured with a system message that provides detailed instructions:

You are a helpful assistant who has access to a bunch of tools to assist with user queries. Before you try to execute any tool, you need to call the tool to list available tools for the capability you want to leverage.

When you list tools available, you'll get a list back of items that look like:

name:[tool_name]
description:[tool description to tell you when and how to use the tool]
schema
0:[param 1]
1:[param 2]
...
n-1:[param n]

Then when you call a tool, you need to give the tool name exactly as given to you, and the tool parameters need to be a json object like:

{
  "param 1": "param 1 value",
  ...
  "param n": "param n value"
}

If there are no parameters for the tool, just pass in an empty object.

For the file system, you have access to the /files directory and that is it.

Setup Requirements

To use this workflow, you'll need:

  1. n8n: A running n8n instance
  2. n8n MCP Nodes: The MCP client nodes for n8n
  3. PostgreSQL: For chat memory persistence
  4. API Credentials:
    • OpenAI API key
    • Brave Search API credentials
    • Convex API credentials

Installation

  1. Import the MCP_Agent_Demo.json file into your n8n instance
  2. Install the n8n community MCP node (instructions there)
  3. Configure the required credentials:
    • OpenAI API
    • PostgreSQL database
    • Brave Search MCP client
    • Convex MCP client

Usage

Once the workflow is activated, you can:

  1. Start a conversation with the agent
  2. Ask questions that require using Brave Search or Convex tools
  3. The agent will automatically discover and use the appropriate tools to answer your questions

Extending the Workflow

You can extend this workflow by:

  1. Adding more MCP clients to provide additional tool capabilities
  2. Customizing the system message to change agent behavior
  3. Adding pre/post processing nodes to enhance the agent's capabilities

Learn More