Self-Hosted n8n + MCP: Build Event-Driven AI Agents (Setup & Demo)

Self Hosted AI Starter Kit: Setup n8n with Ollama & Qdrant for Building AI Agents - https://youtu.be/qdlJk1AeOK0?si=z061DMUdTRrNxDMm

What is MCP (Model Context Protocol)?

Model Context Protocol (MCP) is an open standard that defines how AI agents and language models interact with external tools, data sources, and services.

It enables a structured, event-driven request/response mechanism, allowing models to discover, invoke, and exchange context with external systems.

Key Features:

Standardized Communication: Enables seamless interaction between agents, tools, and LLMs.
Event-Driven: Supports structured request/response over various transports (e.g., stdio, HTTP/SSE).
Modular & Scalable: Easily extend AI capabilities by plugging in new tools via MCP servers.
Interoperable: Designed to work across different AI platforms and environments.
Why MCP?
Standardizes communication between tools, agents, and models.
Facilitates real-time, event-driven AI workflows.
Bridges AI and software with a shared, open protocol.
Enables modular, scalable, and reusable tool integrations.
Works across systems: LLMs, APIs, databases, files, memory, and more.
Simplifies orchestration of multi-step, multi-tool AI agents.

MCP Components
MCP Server:
Exposes tool capabilities to AI agents
Handles event publishing and responses
Example: https://modelcontextprotocol.io/examples

MCP Client
Connects to a specific MCP Server
Sends requests and receives responses

MCP Host (AI application)
Manages one or more MCP clients
Coordinates tool usage and request handling
Examples: Claude, Cursor, Windsurf

MCP Transport Mechanisms

Stdio Transport:

Uses standard input/output (stdin/stdout)
Ideal for local tools or CLI-based agents
Common in dev tools like Cursor or Claude Desktop
Simple and secure for local workflows

HTTP + SSE (Server-Sent Events):

Client → Server: HTTP POST
Server → Client: Server-Sent Events (SSE)
Good for remote tools or services
Enables streamed responses and real-time updates
Often used in hosted AI services

MCP in Action

MCP Host (AI application / orchestrator)

Manages one or more MCP Clients
Decides when to trigger a request
Coordinates tool usage and permission policies
Examples: Claude, Cursor, Windsurf

MCP Client

Acts as the communication layer between Host and Server
Sends structured requests and handles responses
Maintains connection with the designated MCP Server

MCP Server
Interfaces with external tools, services, or data sources
Executes tasks and returns structured results
Exposes available tool capabilities via schema

Demo Agent

User sends a message to our chat agent asking for Google Clander events or create new Events

AI Agent routes the request to an MCP Client node

The Client Connects to a sub-workflow running an MCP Server Tigger

Sub-workflow fetches event data or create new Google Clander Event using the API and responds to the client.

More Details on MCP -https://modelcontextprotocol.io/introduction

MCP support in n8n:
n8n recently introduced support for MCP in its latest versions.

To enable MCP in your self-hosted n8n:

Add this line to your Docker environment config:
N8N_COMMUNITY_PACKAGES_ALLOW_TOOL_USAGE=true

Restart your containers.

Then, from the n8n UI, install the n8n-nodes-mcp package.
This lets your AI Agent node use MCP Client and MCP Server Trigger nodes as tools.

The MCP Client in this case connects to a local MCP server running inside n8n, but it can also connect to external MCP servers.
Check out:
🔗 https://github.com/modelcontextprotocol/servers
for ready-made examples for Figma, GitHub, FireCrawl, and more.

Видео Self-Hosted n8n + MCP: Build Event-Driven AI Agents (Setup & Demo) канала Tech Forum