We're excited to share that the upcoming release will ship with a built-in MCP (Model Context Protocol) server integrated into the OMEdit GUI.
This lets AI assistants — anything that speaks MCP, your own local agents or cloud services — work directly inside your modeling session: reading the active model, editing it, running simulations, and inspecting results, all while you watch it happen in OMEdit.
What the MCP server can do today
The current prototype already supports several common modeling tasks:
- Diagram and icon editing, including connections and drawing shapes.
- Listing and setting a component's parameters.
- Simulation and re-simulation — simulate a class with its default settings, or re-simulate efficiently by tweaking parameters and start-values without rebuilding the executable.
- Awareness of the GUI state — ask which model or plot is currently active, and for multimodal models fetching the content of a digram or plot as an image.
- For when other tools don't exist: reading and writing Modelica code directly.
In practice this means you can ask an assistant things like "add a resistor in parallel with R1 and re-simulate with R2 = 50 Ω" or "minimize overshoot in this PI controller" and watch the changes appear in OMEdit.
The prototype is currently implemented in Linux, but as soon as we get Qt6 fully working on Windows, we'll have that there too.
We want to hear from you
This is the start, not the finished article. Before we lock down the next batch of functionality exposed via MCP, we'd really like to hear how you are thinking about combining AI with your Modelica work:
This is just a start, not the finished article. Before we lock down the next batch of functionality exposed via MCP, we'd really like to hear how you are thinking about combining AI with your Modelica work:
- What workflows do you want to automate or accelerate?
- What functionality would unlock a real use case for you?
- Where do you see AI fitting into teaching, debugging, library development, or industrial modeling pipelines?
Help shape an AI benchmark for Modelica
Alongside the MCP work, we're putting together a benchmark suite of Modelica tasks that an AI can attempt directly and that can be auto-graded. This will allow us to recommend the models that are capable of Modelica modeling, and what parts they excel in.
If you have ideas for tasks worth including — anything from "build this small model from a spec" to "diagnose why this simulation fails" to "tune these parameters to match this reference output" — @sjoelund would love to hear from you by email.
Feel free to comment on this ticket or to open a new one with specific requests.
Your input now will directly shape what ships next.
We're excited to share that the upcoming release will ship with a built-in MCP (Model Context Protocol) server integrated into the OMEdit GUI.
This lets AI assistants — anything that speaks MCP, your own local agents or cloud services — work directly inside your modeling session: reading the active model, editing it, running simulations, and inspecting results, all while you watch it happen in OMEdit.
What the MCP server can do today
The current prototype already supports several common modeling tasks:
In practice this means you can ask an assistant things like "add a resistor in parallel with R1 and re-simulate with R2 = 50 Ω" or "minimize overshoot in this PI controller" and watch the changes appear in OMEdit.
The prototype is currently implemented in Linux, but as soon as we get Qt6 fully working on Windows, we'll have that there too.
We want to hear from you
This is the start, not the finished article. Before we lock down the next batch of functionality exposed via MCP, we'd really like to hear how you are thinking about combining AI with your Modelica work:
This is just a start, not the finished article. Before we lock down the next batch of functionality exposed via MCP, we'd really like to hear how you are thinking about combining AI with your Modelica work:
Help shape an AI benchmark for Modelica
Alongside the MCP work, we're putting together a benchmark suite of Modelica tasks that an AI can attempt directly and that can be auto-graded. This will allow us to recommend the models that are capable of Modelica modeling, and what parts they excel in.
If you have ideas for tasks worth including — anything from "build this small model from a spec" to "diagnose why this simulation fails" to "tune these parameters to match this reference output" — @sjoelund would love to hear from you by email.
Feel free to comment on this ticket or to open a new one with specific requests.
Your input now will directly shape what ships next.