In April 2025 a user asked whether FlexSim would support MCP (thread). The answer was that there are no plans, but that an idea with concrete use cases would be welcome. This is that idea.
What I am asking for: a first-party MCP (Model Context Protocol) server shipped with FlexSim, exposing a documented tool surface over the running application — read the model tree, create and connect objects, create and edit Process Flow activities, read and write labels and properties, run the model, and read statistics. An MCP server is a thin, standard wrapper; the capability it exposes is essentially what FlexScript already does. The value is that any MCP-capable assistant can then work against FlexSim without each of us building and maintaining a private bridge.
Use case 1 — model building through Process Flow. Most of the effort in a new model is not insight, it is assembly: laying out activities, wiring resources and lists, setting parameters, repeating a known pattern for the fifth client. An agent that can create Process Flow activities and connect them lets the engineer describe the logic and then spend their time on the part that actually needs judgement — assumptions, data quality, and interpretation. The engineer still reviews and owns every activity; the agent removes the clicking.
Use case 2 — auditing and documenting existing models. This is the case I would vote for first. Inherited models are a real cost: a model built by someone who has left, with 200 activities, undocumented labels and a Custom Code block nobody wants to touch. An agent able to read the model tree could explain what a flow does, trace where a label is written and consumed, flag unreachable activities or likely deadlock points at resource acquisition, and generate the handover documentation that almost never gets written. Read-only access alone would deliver most of this.
Scope and safety. Read-only tools would already be valuable and are the natural first phase. Write access should be explicit and opt-in, per model or per session, with the server bound locally by default. Nothing here needs to leave the user's machine.
Why first-party matters. Third parties have already built FlexSim MCP servers, which shows the demand is real. But private bridges fragment: each one reverse-engineers the same surface, breaks on version updates, and cannot be shared across the community. An official, versioned tool contract is what makes agent-assisted simulation dependable rather than a demo.
