top of page
WhatsApp Image 2026-10-07 at 11.40.37 PM.jpeg

OpenStudio AI Programming

This YouTube tutorial series from Helix Energy Partners covers advanced OpenStudio methodologies, including building complex custom Ruby/Python measures, interfacing with the OpenStudio MCP, and contributing to the opensource OpenStudio software. The series targets engineers who have mastered OpenStudio fundamentals and are ready to tackle detailed system design, equipment characterization, and programmatic model manipulation.  It may also serve as an entry for University level research projects and thesis via contribution of automation strategies to the Building Component Library (BCL).

Leveraging AI for direct Ruby/Python measures and C++ contributions for OpenStudio/EnergyPlus
October 8th, 2026

Get to Know Us

In these YouTube videos we discuss various strategies for optimizing AI in the building energy modeling and simulation workflow.

 

For power users, steps are included for setting up AI and getting started with contributing to the direct improvement of the open source OpenStudio software via GitHub.

1. AI Coding Tutorial (Part 1 of 2): Using Claude Code to Contribute to OpenStudio on Windows

In this video we walk through the first half of the tutorial (https://drive.google.com/file/d/1IY6asBGGRbcOIFFECRq6Fi1OToht6LCj/view?usp=sharing) for using AI (Claude Code) to help you program and contribute to OpenStudio on a Windows computer, aimed at energy modelers and engineers with little or no programming background. We cover installing the tools, setting up GitHub, forking the OpenStudioApplication repository, and building OpenStudio locally on Windows so you end up with a working, compiled app ready for code changes.

​

  1. Install Claude Code (AI CLI) – Installs Node.js and Claude Code, adds the CLI to your Windows PATH, installs Git for Windows, and authenticates with your Anthropic account.
 
  STOP HERE! If you are only interested in setting up the OpenStudio MCP Server to leverage AI in your building Energy Modeling / Simulations, continue to https://youtu.be/m_ovVINXMd8. 

 

If you want to dig into directly coding the OpenStudio Software to improve the user interface or add features, proceed with the next steps of this tutorial and then proceed to the next video (https://youtu.be/N5rQAw1zqBs).
 

2. Set up GitHub – Creates a GitHub account, configures Git identity and line endings, and installs the GitHub CLI (gh) so you can fork repos and open pull requests from the terminal.   3. Fork the OpenStudioApplication repository – Uses Claude Code and the gh CLI to fork openstudiocoalition/OpenStudioApplication and clone it locally, then creates a CLAUDE.md file so Claude has persistent project context (build command, clang-format version requirements, and known local test failures).   4. Build OpenStudio locally on Windows – Installs the native build toolchain (CMake, Visual Studio Build Tools, Python, Ruby, Conan, Qt), works around a known NREL Conan server SSL issue with a local Ruby package, configures with CMake, and compiles OpenStudioApp.exe. Also covers a Docker/WSL-based alternative build path for anyone who prefers a containerized environment. Timestamps: 0:00 Introduction 4:00 Part 1: Install Claude Code (AI CLI) 5:00 - Install Node.js 5:45 - Install Claude Code 6:40 - Add Claude Code to your PATH 8:10 - Install Git for Windows 9:50 - Authenticate 13:50 Part 2: Set Up GitHub 14:00 - Create a GitHub account 14:40 - Configure Git identity and line endings 20:50 - Install the GitHub CLI 23:30 Part 3: Fork the OpenStudioApplication Repository 23:45 - Fork with AI assistance 32:05 - Create a CLAUDE.md file 33:35 Part 4: Build OpenStudio Locally on Windows 34:11 - Install prerequisites 36:05 - Install Qt 36:30 - Set up Conan 38:05 - Create a local Ruby package for Conan 38:30 - Edit the project's conanfile.py 39:00 - Run Conan install 41:10 - Configure with CMake 43:00 - Build 44:45 - Launch and test

2. OpenStudio EnergyPlus - AI CLI - Setting Up the MCP

In this video we will guide you through setting up the OpenStudio Model Context Protocol (MCP) server, developed by NLR/NREL (National Laboratory of the Rockies), for use with Claude Code (Claude AI). We will install WSL and Docker Desktop, clone and build the MCP server, connect it to Claude Code, and then have Claude use it to modify an OpenStudio model directly.


If you do not already have Claude Code installed and git setup, please follow the instructions on my previous video up through Part 1 here:https://youtu.be/sZu0aEcSVOo


The tutorial md file for this video can be found here: https://drive.google.com/file/d/1ZjtYQWanlf0DzyB01K4xCapLzbgvx3MB/view?usp=sharing


The claude.md file can be found here: https://drive.google.com/file/d/1wx-Dnga9_lQX22qo8uOuMBqHCLR6KmsY/view?usp=sharing

Today we are going to be talking about MCP servers, and specifically the MCP server that was created by the National Laboratory of the Rockies to be used with the OpenStudio application. We will search for it here. It is on GitHub and it is NREL's OpenStudio MCP. What is an MCP? It is a newer thing. It has to do with AI agents and it is a standardized way for AI agents to exchange information. For this specific purpose, it is for the AI agent to understand all of the commands and capabilities of OpenStudio, so that the agent itself can execute those commands and modify your models. If you are cloning an HVAC loop, or injecting a thermal zone, or any of those functions that you do in the OpenStudio application, the AI agent can execute those. In fact, the AI agent has even more capabilities because the OpenStudio application's graphical user interface has only exposed a limited set of the basic needs of the OpenStudio and EnergyPlus software. An AI agent would have the full capabilities of the OpenStudio software development kit. It would have a fuller understanding of how to modify your model. The National Laboratory of the Rockies has developed this MCP server to exchange information with AI agents and basically to teach AI agents how to use OpenStudio. With that, you can basically tell the AI agent in plain English, or whatever your language happens to be, in plain language what exactly you want the AI agent to do. The AI agent can ask you more questions, or if it is simple enough, the AI agent will understand and will query the MCP server to find out what kind of data can be manipulated or changed to modify your model. You could tell the agent: I want to change these windows on the north side of the building to an R-value of this, or a solar heat gain of this. The AI agent can understand that, it can execute the commands for the OpenStudio software development kit, and it can modify your OpenStudio model. With that said, we are going to go through a tutorial on how to get set up with the OpenStudio MCP today. I have created a tutorial. This is an MD file. I will put a link to that in the description below. The idea is that you can go through this yourself, or you can also have your AI go through this with you, and it would help install the OpenStudio MCP. This video presupposes that you have already installed an AI onto your computer. I do go through that in one of my previous videos. I will also put a link to that in the description. That video goes through the installation process for Claude CLI, also known as Claude Code. So that is what we are going to do. Let us go ahead and start Claude. We are going to open up a command prompt and we need to browse to our projects folder. This is a folder that we set up in my previous video. One note about my previous video: there is a lot of extra unrelated stuff, so I would just follow through the steps on that previous video for installing Claude Code. You do not have to continue to the additional steps. I will put a note in the description below. Let us open up that projects folder and let us instantiate Claude. I want to install and test out the OpenStudio MCP. Please guide me through this tutorial, and then we are just going to paste the path to this tutorial. All right. Claude wants to log in. Sometimes that happens. Okay. We are logged in. Claude wants to check if we have WSL, Docker, Git, or a repo or MCP installed. We will just accept. All right. Most of this stuff is not installed. You may have some of these installed, most likely not, but it is entirely possible. Some of this you will have to do by yourself. We need to open up a command prompt separate from Claude. Let us do that. We really need to open up a command prompt with administrator privileges. Let us right click on the Windows button and do Terminal Admin. Click yes. This is PowerShell, but that is okay. Let us copy this and paste it into our PowerShell. Hit enter. It is going to go through installing WSL. WSL stands for Windows Subsystem for Linux and it basically allows you to run Linux on a Windows computer. The reason we need this is because the OpenStudio MCP uses Linux. In order for our Windows machine to talk to the OpenStudio MCP, it uses WSL as an interpreter or a bridge to allow that to happen. Okay, so that is good. We are just going to paste this output back to Claude. Okay. It installed WSL, but Ubuntu was not set up. So first we need to reboot the machine. Let us go ahead and reboot it first. Okay. We are booted back up. Let us open up Claude again, and resume our session. It is going to be this one here. All right. It says: reboot the machine and after it comes up, open up another command prompt and run this right here. Let us copy, right click, Terminal Admin, yes, and paste. We will paste this back to Claude. It says that Ubuntu was not installed. Let us go ahead and copy this and install Ubuntu. Paste this back to Claude. This is one of the issues. We will have to turn off a VPN if we already have one running, as it is not able to resolve this. Let us go ahead and do that now. All right. Let us go ahead and try this again. All right. That works. It is downloading Ubuntu now. If it seems to get stuck, you can cancel and try again. Control-C to cancel. Then let us try to do it again. All right. It is installed successfully and it just wants us to create an account and a password. All right. We will say no for now. Excellent. Ubuntu is installed. All right. We will tell Claude that Ubuntu is installed. It just wants to verify. All right. We have WSL running. The next step is to install Docker Desktop. Docker Desktop is like a sandbox that allows you to run programming code, AI agents, and test programs inside a safe area that is only allowed certain access to certain files. It is contained within this sandbox, so it is not able to get out into your computer. That is what we are going to do here. We will install Docker. We need to go to the website and download Docker Desktop. For us, it is going to be Windows AMD64. Okay, it is downloaded. Let us go ahead and run it. The tutorial says to just go through the regular installation process, but we want to ensure that "Use WSL2 instead of HyperV" is checked. We will click this. Okay. All right. Now that we have installed this new software, we need to close and restart. We are back and we have restarted. It is just going through the service agreement and we are just going to hit Accept. We can skip this stuff for now. Let us open Claude back up. Okay. Docker is running. Claude is going to verify the version. Claude is just going through some setup notes that I have in the tutorial. Now Claude is going to clone the OpenStudio MCP repository. Click yes. This is getting all of the latest information from the MCP repository. Now it is going to build the Docker image. It is building the MCP inside the Docker that we just created. Click yes. Creating directories. Right now it is just building the MCP inside the Docker for us. It is basically building a mini computer in our sandbox that we can use to modify our OpenStudio models in that safe environment. Okay, there is a light power failure here. Let us do this. The next step is to restart Claude, because this session of Claude does not know about or is not connected to the MCP yet. Let us exit out of this and re-enter. Okay. We will type /mcp. It says the OpenStudio MCP is now connected. These are all the tools that the MCP has that Claude has access to. All right. We will hit Escape. Escape. Escape. Claude is asking for the version. Great. It is fully working and now we can jump in and start asking Claude to do something. What we are going to do is something simple. I have an example model here. We will just do something simple, like for instance, this gas heating coil here. This will maybe be easier doing it ourselves, but this is just a demonstration of Claude doing it for us. We have this coil heating gas here and I am just going to ask Claude to change this to an electric heating coil instead. We will copy the path. I have this model at this location. I want to change the coil heating gas one to an electric heating coil instead. What I am going to do here is just let Claude run free with it. Right now we are on manual mode with Claude and I am just going to do Shift-Tab until we get to automatic. That will turn Claude loose and it can just hammer away at it. Click yes. Shift-Tab. Auto. All right. Claude has successfully...well, we will see if it successfully did this, but it looks like it had to write a custom measure, a custom piece of code, to replace that coil for us. One of the things with the OpenStudio MCP is that it has a lot of connectors that allow you to do these things, but it does not actually have methods to do some of this stuff. That is not a problem, though, because AI is really good at creating custom programs. AI can figure out how to make those connections work. That is what Claude had to do here. I will talk about this in some of my future videos as far as adding a code base so that Claude has code snippets it can use. For now, let us go ahead and take a look at what Claude did and see how it turned out. I am going to have it overwrite this model for us. Let us go ahead and take a look at the model. We have got this gas heating coil here, and we asked it to replace it with an electric coil. Let us go to File, Revert to Saved. That should open up the new model that Claude just overwrote. Yes, here it is. It was successfully able to replace that gas coil with an electric heating coil. This was just a simple example. Claude can do a lot of this simple stuff, and more complicated stuff too. I go into some of my other videos about writing custom measures, and AI agents are really good at writing code. That is probably what you will end up using this for most of all: to write custom programs to modify your model, then inspect the energy use, loads, and simulating conditions. That is it in a nutshell for installing the OpenStudio MCP. I hope this tutorial was helpful for you. I look forward to seeing you in the future. Thank you. Please like and subscribe.

3. AI Coding Tutorial (Part 2 of 2): Using Claude Code to Contribute to OpenStudio on Windows

In this video we walk through the second half of the tutorial (https://drive.google.com/file/d/1IY6asBGGRbcOIFFECRq6Fi1OToht6LCj/view?usp=sharing) for using AI (Claude Code) to help you program and contribute to OpenStudio on a Windows computer, aimed at energy modelers and engineers with little or no programming background. Picking up with a working, compiled build from Part 1, we create a branch, use Claude Code to find and fix a real reported issue, rebuild and test the change, and submit it as a pull request. We close with practical tips learned the hard way from a real setup and contribution.

1. Create a branch and solve an issue with AI – Reads and understands a real GitHub issue (Issue #857), creates a new branch off develop, works with Claude Code to locate and fix the relevant code, rebuilds incrementally, tests the fix in the running app, commits the change (without staging machine-specific build workarounds), and pushes the branch to open a pull request. 2. Tips from real experience – Practical lessons for working with Claude Code day to day: treating it as a collaborator rather than a magic button, keeping CLAUDE.md updated, using "!" to run shell commands without leaving a session, fixing pre-commit hook failures, running Ruby scripts through OpenStudio's bundled interpreter, handling SSL errors safely, and asking better questions to get better results. Timestamps: 0:00 Introduction 1:15 - Understand the issue 9:30 Part 5: Create a Branch and Solve an Issue With AI 9:45 - Create a new branch 11:14 - Work with Claude to find and fix the code 12:45 - Rebuild incrementally 13:15 - Test your fix 21:30 - Commit your changes 23:11 - Discuss Push and create a pull request 23:48 - Work with Claude to solve the second part of the issue 37:15 - Rebuild incrementally 37:33 - Test your fix 40:45 - Commit the changes 42:15 - Try to Push with new Github account - Unsuccessful 43:30 - Create a whole new Repository with the fix 47:53 - Try to fork on github.com - Unsuccessful 49:08 - Summary

bottom of page