Agentic AI Development using n8n and MCP is a course in building AI agents with n8n, from ReAct and Reflexion patterns to self-hosting n8n with Docker, connecting MCP servers to Slack and MSSQL and building RAG with Qdrant. It suits developers, existing n8n users and IT teams running self-hosted systems, and you leave with a customer service agent that answers from real company data.
Traditional automation follows conditions written in advance, so it stalls on a message that does not fit the pattern or a question that needs data from several places. Agentic AI changes this by letting an LLM interpret the request, plan, and choose which tools and data sources to use before it answers or acts. The real challenge is connecting that LLM to an organization's email, chat, databases and documents in a way that works reliably and stays under control.
This course uses n8n as the backbone for assembling AI agents and the Model Context Protocol (MCP) as the standard way for agents to reach tools and data. Learners start with the concepts of agentic AI and reasoning patterns such as ReAct, Reflexion and Chain-of-Thought, then build an AI email classification and reply system on n8n Cloud. They move on to self-hosting n8n Community Edition
with Docker, connect an MCP server to Slack and a Microsoft SQL Server database, and build a RAG system with Qdrant. Everything comes together as a customer service agent that answers from real transaction history and policy documents, handles multi-turn conversations, hands off to a human when needed, and is ready to scale into production. (3 days, 6 hours per day, 18 hours in total, Intermediate level.)
Course Details
A 3-day course, 6 hours per day (18 hours in total, 09:00-16:00), delivered as lectures with labs that follow one continuous use case across all three days. Intermediate level. Day one runs on n8n Cloud; days two and three run on each learner's own machine with Docker. Learners bring their own Gmail and Slack accounts and an OpenAI API key with credit; OpenAI
API usage is billed by actual use. The course focuses on building AI agents with n8n and MCP rather than on workflow automation basics. Anyone who wants a solid n8n foundation first should take the separate n8n Workflow Automation for Modern Business course. Learners take home a lab guide, n8n workflow files for every lab, and docker-compose files for n8n, the MCP server, MSSQL and Qdrant.
Section 1: Agentic AI and How Agents Reason
What agentic AI means and how it differs from rule-based automation
Core components: perception, planning, reasoning, memory and action
ReAct: the agent alternates reasoning with tool calls until it reaches an answer
Reflexion: the agent reviews its own output and corrects it
Chain-of-Thought, and choosing a pattern that fits the task and the cost
Real use cases in customer service, document work and operations
Section 2: The n8n Architecture
Low-code / no-code workflow automation and where n8n fits
Core building blocks: nodes, connections, workflows and credentials
Runtime architecture and how workflows are executed
n8n Cloud compared with self-hosting
Community Edition compared with Enterprise Edition
Section 3: Key Nodes for Agentic AI
Trigger nodes: Webhook, Gmail and Schedule
AI nodes: OpenAI and Text Classifier for categorising messages
Communication nodes: Gmail, Slack and HTTP Request
Data processing nodes: Code, JSON and Split in Batches
Database nodes (MSSQL, MongoDB) and utility nodes (IF, Switch, Merge)
Section 4: Lab: n8n Cloud and AI Email Classification
Sign up for and configure n8n Cloud, with good practice on the cloud platform
Connect Gmail and OpenAI credentials securely
Configure a Gmail trigger to bring new email into the workflow
Use Text Classifier to sort email into work, personal, travel and other
Use IF / Switch nodes to route each category down a different path
Section 5: Lab: Automated Email Replies and Scheduling
Connect an OpenAI model to the email classification workflow
Write prompts that draft replies with the right context and tone
Configure the Gmail node to send replies automatically
Create Google Calendar events from the content of an email
Test the whole workflow and check results in the execution log
Section 6: Lab: Installing n8n Community Edition with Docker
Prepare the machine and install Docker and Docker Compose
Write and customise a docker-compose.yml file for n8n
Run n8n and open its web interface on your own machine
Configure environment variables and store credentials securely
Move workflows from n8n Cloud to the self-hosted instance
Section 7: Lab: MCP Server with n8n
MCP architecture: host, client, server and exposing tools to an agent
Install an MCP server with Docker and configure it
Connect the MCP server to n8n so the agent can call its tools
Test the MCP server and inspect the results of tool calls
Section 8: Lab: Connecting Slack to the Workflow
Create a Slack app and obtain an API token with only the scopes needed
Configure webhooks and Event Subscriptions so n8n receives messages
Connect Slack to an n8n workflow
Test sending and receiving messages through the Slack API
Section 9: Lab: MSSQL Database through MCP
Install Microsoft SQL Server on Docker and load sample customer data
Configure the database connection through the MCP server
Have the agent retrieve customer transaction data and test the results
Limit the database account so the agent can reach only what it needs
Section 10: Lab: A RAG System with Qdrant
The idea behind retrieval-augmented generation and why it helps question answering
Install the Qdrant vector database on Docker
Load customer service policy documents into Qdrant with OpenAI embeddings
Build the retrieval component and connect RAG to the MCP server
Test searching for the policies relevant to a customer's question
Section 11: Designing the Customer Service Agent
Design the user flow and conversation sequence with the customer
Context management so the agent knows who it is talking to and about what
Designing the system prompt and prompts that keep answers within policy
Planning data handling and how answers are built from several sources
Section 12: Lab: Building the Customer Service Agent
Build a workflow that receives Slack messages and lets the agent interpret them
Retrieve the customer's transaction history from MSSQL through MCP
Search Qdrant for the relevant policies
Generate an answer from customer data and policy, then reply in Slack
Log the conversation history to MSSQL for later reference
Section 13: Lab: Handling Complex Conversations
Support multi-turn conversations
Strategies for retaining context across messages
Fallback when the agent is unsure, and handoff to a human agent
Refine prompts so answers follow policy consistently
Section 14: Performance Tuning and Scaling
Analyse performance and find bottlenecks in the workflow
Tune the Docker configuration for production
Scaling strategies for n8n and the MCP server as load grows
Guidance for taking the agent into real organisational use
ReAct (reasoning and acting) has the agent alternate between thinking and doing: decide what to do, call a tool, observe the result and reason about the next step until it has an answer, which suits tasks that need several lookups or system calls. Reflexion has the agent review its own output afterwards, find mistakes and apply the lessons in the next attempt, improving quality at the cost of more time and tokens. Choose according to task complexity and budget.
Install Docker and Docker Compose on the machine or server, then write a docker-compose.yml that uses the n8n image, publishes port 5678, mounts a volume for n8n's data folder so workflows and credentials survive when the container is recreated, and sets environment variables such as N8N_ENCRYPTION_KEY and the timezone. Run docker compose up -d and open localhost:5678 in your browser.
First build an ingestion workflow that splits the policy documents into chunks, converts them into vectors with OpenAI embeddings and stores them in a Qdrant vector store. Then connect Qdrant to the AI agent workflow as a retrieval tool, so when a customer asks a question the agent fetches the closest passages to build its answer. Instruct it in the system prompt to answer only from retrieved documents and to say so when nothing is found.
It suits developers who want AI agents connected to real company systems, existing n8n users moving from automation to agentic AI, business analysts and project managers designing automated customer service, and IT teams running self-hosted n8n. You should be comfortable with computers and understand the basics of APIs, webhooks and JSON. Light coding experience helps, but you do not need to be a programmer.
You will be able to install n8n Community Edition with Docker Compose and manage credentials securely, build workflows that sort and answer email and book meetings in Google Calendar, connect an MCP server so agents can query MSSQL, integrate Slack, and build a customer service agent that answers from documents with RAG on Qdrant and hands over to staff when needed. Participants who complete the training receive a Certificate of Completion with its own number, which can be verified online.
Bring Gmail and Slack accounts for testing, an OpenAI API key with credit, and a laptop that can run Docker Desktop with permission to install software, as the later labs run n8n, an MCP server, MSSQL and Qdrant on your own machine. You take home the lab guide, workflow files and docker-compose files for every lab.
Repetitive tasks such as copying form data into spreadsheets, sending confirmation emails, compiling daily totals or alerting the team about new orders take up a large share of staff time every week. Workflow automation lets systems hand work to each other…
AI today is more than a chatbot that answers questions. Agentic AI can plan and carry out work for us, built around two open standards: MCP (Model Context Protocol), which gives AI a single standard way to use tools, files and databases, and A2A…
General AI models are strong at language but do not know your organization's internal information - manuals, policies, contracts, customer data or accumulated know-how. When asked something specific to the organization, AI often answers incorrectly…
