AI and machine learning
AI Agents and MCP for Engineers
Build AI agents in Python that read plant data safely, using MCP servers for a SQL historian, OPC UA and Modbus.
AI Agents and MCP for Engineers is a free, self-paced online course from EDWartens for automation, controls, maintenance, reliability and data engineers at plants, system integrators and EPC contractors worldwide who want AI agents that read plant data safely. It has 13 modules and takes about 17.5 hours, with written notes, a practice task per module, a practical project and a final assessment of 15 questions (pass mark 60%). Learning costs nothing and needs only a free account. The optional EDWartens Certificate of Completion, verifiable by code, is a one-off US$28.99 for you; it is not a vendor credential. Updated 1 November 2026.
- Free
- Intermediate
- 17.5 hours
- Self-paced
- Certificate optional
- Lab practice in your browser
No card, no trial. The course is free for good. Certificate US$28.99, optional.
- Who it is for
- Automation, controls, maintenance, reliability and data engineers at plants, system integrators and EPC contractors worldwide who want AI agents that read plant data safely
- Jobs
- Roles that use this skill: AI engineer, machine learning engineer, data analyst and automation developer. No course guarantees a job.
- Tools
- Python 3.10 or later with FastMCP, asyncua, pymodbus, the OpenAI Agents SDK and LangGraph (all free, installed with pip), VS Code, SQLite, and at least one MCP host: Claude Desktop or Claude Code, VS Code with GitHub Copilot, or a local model in Ollama. Every exercise can be done with free tools and a local model. Hardware: A computer with 16 GB of memory is enough for a small local model; 8 GB works if you use a hosted model instead. No PLC or plant equipment: every exercise runs against simulators and a sample historian.
- Updated


After this course you can
By the end you will be able to:
- Explain what an AI agent is and when a plain workflow is better
- Write tool definitions and a tool-calling loop in Python
- Build read-only MCP servers with FastMCP
- Connect MCP servers to Claude, VS Code and a local model
- Build a maintenance triage agent that stops for human approval
Who it is for, and what you need first
Who it is for
Automation, controls, maintenance, reliability and data engineers at plants, system integrators and EPC contractors worldwide who want AI agents that read plant data safely.
Level: Intermediate. Intermediate. For automation, controls, maintenance and data engineers who can read and run a short Python program. You do not need to have built an AI application before; the RAG and n8n courses are useful but not required.
What you need
Software: Python 3.10 or later with FastMCP, asyncua, pymodbus, the OpenAI Agents SDK and LangGraph (all free, installed with pip), VS Code, SQLite, and at least one MCP host: Claude Desktop or Claude Code, VS Code with GitHub Copilot, or a local model in Ollama. Every exercise can be done with free tools and a local model..
Hardware: A computer with 16 GB of memory is enough for a small local model; 8 GB works if you use a hosted model instead. No PLC or plant equipment: every exercise runs against simulators and a sample historian.. What to download, and how.
How this course works
Watch the lessons
Short video lessons, one module at a time, at your own pace.
Read the notes
Open the Reading tab for the written notes and diagrams of the module.
Practise in the 3D labOptional
Try the module in the free 3D lab, in your browser.
Check yourself
Answer the questions to see what stuck before you move on.
Final assessment
Pass the final once you have been through the modules.
Certificate, if you want oneOptional
Optional, with a page anyone can use to verify it.
What you learn
13 modules in order, at your own pace. 17.5 hours of video lessons, with written notes and a practice task in each module.
- 1What an agent is: model, tools, memory and stop conditions31m
- 2Tool calling from Python, with lab practice: 40m
- 3The Model Context Protocol: hosts, clients, servers and primitives1h 21m
- 4Your first MCP server with FastMCP1h 3m
- 5An MCP server for a SQL historian, read only by design, with lab practice: 1h 10m
- 6MCP for OPC UA and Modbus: browse, read, and why writes stay off, with lab practice: 1h 0m
- 7Using MCP servers from Claude, VS Code and local models1h 1m
- 8Agent frameworks compared: OpenAI Agents SDK, LangGraph, Microsoft Agent Framework1h 28m
Show the other 5 modules
- 9A multi-step maintenance triage agent, with lab practice: 26m
- 10Guardrails: approval, allow-lists, rate limits, audit logs and prompt injection, with lab practice: 1h 0m
- 11Evaluating agents: task success, traces, cost and latency26m
- 12Deploying agents for a team: containers, secrets and network zones1h 0m
- 13Final Assessment
Software you need
What to download, where from, what it costs and how to install it. Every link goes to the maker's own site, never a mirror.
FastMCP, asyncua, pymodbus, LangGraph and the OpenAI Agents SDK install with pip; SQLite comes with Python. Claude Desktop runs local MCP servers from its config file. Node.js is only needed for the MCP Inspector, and Docker only for the deployment module.
Show what to download (11 tools)
Required
- 01Free
Python
Python Software Foundation
- Runs on
- Windows (not Windows 7 or earlier), macOS and Linux. Windows builds for x64, 32-bit and Arm64.
- Account
- None needed
Python is free, open source software. You can use it for learning and for commercial work at no cost.
Steps
- 1.Open python.org/downloads and click the download button for your system.
- 2.On Windows, run the Python install manager (or the classic 64-bit installer) you downloaded.
- 3.If you use the classic installer, tick "Add python.exe to PATH" on the first screen, then click Install Now.
- 4.Open a new Command Prompt or Terminal and run: python --version (on Windows you can also run: py --version).
- 5.Install packages with pip, for example: python -m pip install requests
- pip comes with Python. Run it as python -m pip so it always matches the Python you are using.
- Python.org now recommends the Python install manager on Windows. If it offers to add its folder to PATH, say yes so the python command works everywhere.
- Make a virtual environment for each project: python -m venv .venv
Official download pagepython.orgAlternatives
- Anaconda Distribution: Python with 600+ data science packages and Jupyter already included.
- 02Free
Visual Studio Code
Microsoft
- Runs on
- Windows 64-bit (supported Windows client versions), macOS (latest and two previous releases), Linux (Ubuntu 20.04, Debian 10, RHEL 8, Fedora 36 or later)
- Account
- None needed
- Size
- Less than 200 MB download, under 500 MB installed
Free to download and use. Extensions from the Marketplace each have their own licence.
Steps
- 1.Open code.visualstudio.com/download.
- 2.Pick the installer for your system (on Windows, the User Installer is fine).
- 3.Run the installer. On Windows, tick "Add to PATH" if it is offered.
- 4.Open VS Code and install the extensions your course uses, for example the Python extension.
- Hardware needs are small: a 1.6 GHz processor and 1 GB of RAM.
- On Windows, the "Open with Code" options in the installer let you open folders from File Explorer.
Official download pagecode.visualstudio.com
One MCP host or local model
Any one is enough.
- 03Free plan with limits
Claude
Anthropic, in the browser
- Runs on
- Any modern web browser, plus desktop and mobile apps
- Account
- A free Claude account (you must be 18 or over)
The Free plan covers chat on web, desktop and mobile, web search, file creation and Artifacts. Usage limits reset on a rolling five-hour window and depend on message length and features used. Paid plans give more usage.
Steps
- 1.Open claude.ai (or see the plans at claude.com/pricing).
- 2.Sign up with your email or Google account and confirm your email.
- 3.Confirm you are 18 or over and accept the terms.
- 4.Type your question in the message box.
- Long conversations and large files use up your limit faster. Start a new chat for a new topic.
- Do not paste personal data or confidential work files.
Open Claudeclaude.com - 04Paid plan or API account
Claude Code
Anthropic
- Runs on
- macOS 13 or later, Windows 10 1809 or later, Ubuntu 20.04 or later, Debian 10 or later; 4 GB of RAM
- Account
- A Claude Pro, Max, Team or Enterprise plan, or a Claude Console (API) account
The free claude.ai plan does not include Claude Code. It runs in the terminal, in VS Code and JetBrains through extensions, and in a desktop app.
Steps
- 1.Open code.claude.com/docs/en/setup and follow the native installer for your system.
- 2.On macOS, Linux or WSL run: curl -fsSL https://claude.ai/install.sh | bash. On Windows PowerShell run: irm https://claude.ai/install.ps1 | iex.
- 3.Open a new terminal and run claude --version, then claude doctor.
- 4.Go to your project folder, run claude and follow the browser prompt to log in.
- 5.Run /init to create a CLAUDE.md, then cut it down to the rules that matter.
- On native Windows, install Git for Windows so Claude Code can use Git Bash.
- Start new work in plan mode (Shift+Tab) and commit before every task.
Anthropic product pagecode.claude.comLegitimate free routes
- Cursor (Hobby plan): A free plan with limited Agent requests
- GitHub Copilot Free in VS Code: A free monthly allowance
- 05Free plan with a monthly allowance
GitHub Copilot in VS Code
GitHub (Microsoft)
- Runs on
- Wherever VS Code runs: Windows, macOS and Linux
- Account
- A GitHub account
Copilot Free gives a monthly allowance of inline suggestions and AI credits. Verified students can use Copilot Student. Paid plans (Pro, Pro+, Max, Business, Enterprise) add more usage and Copilot cloud agent; check GitHub's plans page for the current terms.
Steps
- 1.Install VS Code from code.visualstudio.com/download.
- 2.Open VS Code, select the Copilot icon in the Status Bar and choose Use AI Features.
- 3.Sign in with your GitHub account when the browser asks.
- 4.Open the Chat view and pick Agent or Plan for a task in your project.
- 5.Add a .github/copilot-instructions.md or an AGENTS.md with your project's rules.
- Copilot cloud agent is not part of Copilot Free; on Business and Enterprise an administrator must enable it.
Official download pagecode.visualstudio.com - 06Free
Ollama
Ollama Inc.
- Runs on
- macOS 14 Sonoma or later, Windows 10 or later, Linux
- Account
- None needed
Ollama is free and open source (MIT licence). Each model you download has its own licence; check it before commercial use.
Steps
- 1.Open ollama.com/download and pick your system.
- 2.On Windows run OllamaSetup.exe; on macOS open the DMG; on Linux run: curl -fsSL https://ollama.com/install.sh | sh
- 3.Open a terminal and run a small model, for example: ollama run <model name> (browse names on ollama.com).
- 4.Type a question at the prompt. Type /bye to exit.
- Models are several GB each. Check your free disk space first.
- More RAM, or a supported GPU, lets you run larger models faster.
- On Windows you can also install from PowerShell: irm https://ollama.com/install.ps1 | iex
Official download pageollama.com
Optional
Useful, not needed to finish the course.
- 07Free
Git
Git project (git-scm.com)
- Runs on
- Windows (x64 and ARM64), macOS, Linux
- Account
- None needed
Git is free and open source software. Use it for any purpose at no cost.
Steps
- 1.Open git-scm.com/downloads and choose your system.
- 2.On Windows, download the x64 Setup (or ARM64 Setup) and run it. The default choices are fine for most learners.
- 3.Open a new terminal and run: git --version
- 4.Set your name: git config --global user.name "Your Name"
- 5.Set your email: git config --global user.email "you@example.com"
- On Windows you can also install with: winget install --id Git.Git -e --source winget
- A portable version (no installer) is offered for PCs where you cannot install software.
Official download pagegit-scm.com - 08Free
Node.js
OpenJS Foundation
- Runs on
- Windows, macOS and Linux
- Account
- None needed
Free, open source runtime. npm, the package manager, is installed with it.
Steps
- 1.Open nodejs.org/en/download.
- 2.Choose the LTS (long-term support) version and your system.
- 3.Download and run the installer with the default options.
- 4.Open a new terminal and run: node -v and then npm -v
- Pick LTS, not Current, unless your course says otherwise.
- Some tools need a minimum Node.js version. Node-RED 5 needs Node.js 22 or later.
Official download pagenodejs.org - 09Free for personal use, education and small businesses
Docker Desktop
Docker, Inc.
- Runs on
- Windows 10 64-bit version 22H2 (build 19045) or Windows 11 64-bit version 23H2 (build 22631) or later, with WSL 2 and hardware virtualisation on, 8 GB RAM. Also macOS and Linux.
- Account
- None needed
Free for personal use, education, non-commercial open source projects, and businesses with fewer than 250 employees and less than 10 million US dollars in annual revenue. Larger businesses need a paid subscription. Docker Engine on Linux is open source and is not covered by these terms.
Steps
- 1.Open docker.com/products/docker-desktop and download the installer for your system.
- 2.On Windows, install WSL first (wsl --install) if you do not have it.
- 3.Run the installer and keep the WSL 2 option selected.
- 4.Restart if asked, then start Docker Desktop and accept the subscription terms.
- 5.Open a terminal and run: docker run hello-world
- Turn on virtualisation in BIOS or UEFI if Docker says it is not available.
- On Linux servers, install Docker Engine instead of Docker Desktop.
Official download pagedocker.comAlternatives
- Docker Engine (Linux): Free, open source engine for Linux. Not covered by the Docker Desktop subscription terms.
- 10Free
PyModbus
pymodbus-dev (open source project), a Python or npm package
- Runs on
- Windows, macOS and Linux with Python 3.10 or later
- Account
- None needed
Free, open source under the BSD 3-Clause licence.
Steps
- 1.Install Python 3.10 or later and create a virtual environment.
- 2.Run: pip install "pymodbus==3.8.6" (the version the course code is written and tested for; 3.10 renamed slave= to device_id=)
- 3.For serial (RS-485) devices, run: pip install "pymodbus[serial]==3.8.6"
- 4.Follow the client examples in the documentation to read holding registers from a device or the built-in simulator.
- PyModbus includes a server simulator with a web interface, so you can practise without real hardware.
- Only write registers on lab equipment. Writes can move real machines.
Official download pagegithub.com - 11Free
LangChain
LangChain, Inc., a Python or npm package
- Runs on
- Any system with Python 3.10 or later
- Account
- None needed
Free, open source under the MIT licence. Model providers you call may charge for use.
Steps
- 1.Install Python 3.10 or later and create a virtual environment.
- 2.Run: pip install -U langchain
- 3.Install the package for your model provider, for example: pip install -U langchain-openai or pip install -U langchain-anthropic
- 4.Set your provider's API key as an environment variable and run the quickstart example from the docs.
- Keep API keys in environment variables or a .env file, never in shared code.
Official download pagedocs.langchain.com
Checked against each maker's own page on 27 September 2026. Trial lengths and editions change; the maker's page is the final word.
Practise in the 3D lab
Practise the guardrails behind a plant agent in the AI Workbench: tool calls within limits, approval before action, citing evidence and resisting planted instructions.
- Answers with sourcesAI Workbench, about 25 minOpen in the lab (opens in a new tab): Answers with sources
- A safe agentAI Workbench, about 30 minOpen in the lab (opens in a new tab): A safe agent
Optional. Skip it whenever you like: it never blocks your progress or your certificate.
Course project, about 14 hours
A read-only trip-analysis agent for a raw water pump: MCP servers, approval and an evaluation report
Build an agent that answers 'why did pump P-101 trip?' with cited evidence and drafts a work order that a person approves. You write read-only MCP servers for live OPC UA tags from a simulator, the historian and the maintenance log, add guardrails, and evaluate the agent on fixed cases. Then you place it in the plant's zones. The sample pack shows each document for a municipal water treatment works.
- 5documents in the pack
- 8steps to follow
- Laboptional: in the 3D lab
- 8things to hand in
- 14 hof work, at your pace
- FDSFunctional specification: P-101 trip-analysis agent
- Tag listTag list: tags the OPC UA server may read
- Zones & conduitsZone and conduit register: trip-analysis agent
- Risk registerRisk register: trip-analysis agent
- Test reportEvaluation report: trip-analysis agent, version A
Do it in the lab: 5 steps.
The project is optional practice. The certificate needs only the modules and the final assessment.
After the course, your next step
Show what you can do, then keep going in the same field.
A certificate you can verify
Finish the modules and pass the final assessment (15 questions, pass mark 60%) and you can add a certificate with a public verification page. One-off, US$28.99 for you. It is not a vendor credential.
Want a free certificate? Do the matching Free internship: 4 weeks, nothing to buy, with a free certificate of completion.

Generative AI and Prompt Engineering
Eight short weeks of how language models work, prompting that gets better answers, and the risks to manage: notes, diagrams, a few videos and worked exercises. Finish weeks 1 to 4 and the final, then claim a free Internship Completion Certificate with your real dates and study hours, and a QR code anyone can scan to verify it. Training, not employment.
See the internshipMore free courses
Roles that use this skill
- AI engineer
- Machine learning engineer
- Data analyst
- Automation developer
These are the kinds of roles where the skill is used. We make no promise of a job or an interview.
Plain notes
- Free: the whole course, the notes, the practice tasks, the final assessment and support from practising engineers.
- Paid, and optional: the course certificate, and the Engineer-verified internship (the Free internship costs nothing).
- No job promise: the course teaches a skill. It does not place you in a job.
- Lessons by: IBM Technology, Anthropic, freeCodeCamp.org, Microsoft Reactor, Tech With Tim, Industry40tv, OpenAI, LangChain, Visual Studio Code, Docker, Google Cloud Tech, AAIF Live. Independent creators; the lessons stream from their own channels. EDWartens wrote the study plan, notes, practice tasks and assessments.
IBM Technology · Anthropic · Claude · Dave Ebbelaar · Mervin Praison · Google Cloud Tech · freeCodeCamp.org · Microsoft Reactor · Tech With Tim · Industry40tv · APMonitor.com · IoT Frontier · Fusion Automate · Visual Studio Code · OpenAI · LangChain · N3uron · SAS FSA · AAIF Live · Langfuse · Docker · Automation World · ByteByteGo. If you are one of these creators and want a lesson removed or credited differently, write to info@edwartens.com.
Practise what you learn in the free 3D lab
Technical work is judged on what you can do: wiring, commissioning, finding a fault. Doing a task yourself teaches more than watching it a second time, and a lab lets you try it safely before you meet real equipment. It does not replace working on real hardware.
- Practise safely, so a mistake costs nothing.
- The lab checks your work like a site test, so you know when it is right.
- Free, in your browser, with nothing to install.
Why learn with EDWartens
What you get with this course, and what comes after it.
- Free from the first lesson to the final assessment.
- Practice in a free 3D lab in your browser, with nothing to install.
- Video lessons with written notes and a project.
- An optional certificate that anyone can verify online.
- Free internships to keep practising afterwards.
Common questions
Who is this AI agents and MCP course for?
It is for automation, controls, maintenance, reliability and data engineers who want AI agents that read plant data safely, including engineers at plants, system integrators and EPC contractors worldwide. You should be able to read and run a short Python program.
What is MCP and why does an engineer need it?
The Model Context Protocol is an open standard for connecting AI hosts and agents to tools and data. It is now governed by the Agentic AI Foundation under the Linux Foundation. Wrap a historian, an OPC UA server or a maintenance log once as an MCP server. Then Claude, VS Code or your own agent can use it without a custom connector each time.
Is the course free to learn, and what software do I need?
Yes. Every module, the notes, the code, the project and the final assessment are free to learn. The software is free too. You need Python with FastMCP, asyncua, pymodbus, LangGraph or the OpenAI Agents SDK. You also need either a free MCP host or a local model in Ollama.
Will an agent from this course write to my PLC?
No. Every tool that touches plant data is read only by design. Every exercise runs against simulators and a sample historian. Anything with consequences stops at a draft that a person approves. Agent actions are never run on a live plant without review and testing.
How is this different from the n8n, RAG and AI coding courses?
AI Agents and Workflow Automation with n8n builds agents without code. RAG and Chatbots with LangChain covers answering from documents. AI Coding for Engineers covers coding agents such as Claude Code. This course is code-first and industrial. You build MCP servers and agents in Python over historian, OPC UA and Modbus data, with guardrails and evaluation.
How long does the AI Agents and MCP for Engineers course take?
About 17 hours of video, notes and practice at your own pace, of which about 12.6 hours is video. The optional trip-analysis agent project takes about 14 hours more.
What certificate does the AI Agents and MCP for Engineers course give?
An EDWartens Certificate of Completion, issued when you pass the 15-question final at 60 percent, with a number anyone can verify on our site. It is not a certification from Anthropic, OpenAI, Microsoft, LangChain or the Agentic AI Foundation.
What job does this course lead to?
It builds the skills of an industrial AI or automation software engineer. You connect AI agents to plant systems through MCP, with the guardrails, evaluation and OT security placement that plants and integrators ask for. To deepen the security side, go on to AI Security and the OWASP Top 10 for LLMs and ICS Security Foundations.
Do I have to use the lab?
No. The lab is optional practice. You can skip it whenever you like, and it never blocks your progress, the final assessment or your certificate.
Is the lab free?
Yes. Every lab level for this course opens in your browser, free, with nothing to install.
What topics does the AI Agents and MCP for Engineers course cover, module by module?
The AI Agents and MCP for Engineers course has 13 modules. It opens with "What an agent is: model, tools, memory and stop conditions", goes through "Tool calling from Python", "An MCP server for a SQL historian, read only by design", "A multi-step maintenance triage agent" and "Deploying agents for a team: containers, secrets and network zones" and ends with "Final Assessment". The full list with minutes is under "What you learn" on this page.
What can I do after the AI Agents and MCP for Engineers course?
After the AI Agents and MCP for Engineers course: Explain the agent loop and choose between a workflow and an agent; Write tool definitions and a tool-calling loop in Python, including against a local model in Ollama; Describe MCP hosts, clients, servers, tools, resources and prompts as of the 2026-07-28 specification. These are the practical outcomes the course is written around, and the final assessment of 15 questions checks them.
What software and hardware do I need for the AI Agents and MCP for Engineers course, and does it cost anything?
For the AI Agents and MCP for Engineers course, software: Python 3.10 or later with FastMCP, asyncua, pymodbus, the OpenAI Agents SDK and LangGraph (all free, installed with pip), VS Code, SQLite, and at least one MCP host: Claude Desktop or Claude Code, VS Code with GitHub Copilot, or a local model in Ollama. Every exercise can be done with free tools and a local model. Hardware: A computer with 16 GB of memory is enough for a small local model; 8 GB works if you use a hosted model instead. No PLC or plant equipment: every exercise runs against simulators and a sample historian. The course itself is free.
Is the AI Agents and MCP for Engineers course suitable for me, and what level is it?
AI Agents and MCP for Engineers is rated intermediate. Intermediate. For automation, controls, maintenance and data engineers who can read and run a short Python program. You do not need to have built an AI application before; the RAG and n8n courses are useful but not required. It is written for automation, controls, maintenance, reliability and data engineers at plants, system integrators and EPC contractors worldwide who want AI agents that read plant data safely, and takes about 17.5 hours at your own pace.
Which jobs use the skills from the AI Agents and MCP for Engineers course?
Skills from the AI Agents and MCP for Engineers course (for example: explain the agent loop and choose between a workflow and an agent) are used in roles such as AI engineer, machine learning engineer, data analyst and automation developer, and in industrial AI engineer work. The course gives you practical evidence of the skill; it does not promise a job, and employers decide on experience and interview.
Start free today
AI Agents and MCP for Engineers, at your own pace. Add the certificate only if you want one.