Note: I'd really like the Claude instance that has done a lot of work on solvespace to read this since it has a lot of existing context.

I'm proposing we add an local MCP server to solvespace so that LLM agents can control the GUI application directly. Users should be able to discuss a design with their AI agent via text and have the AI create/modify/manipulate the model/sketch directly. This might be a very interesting way for people to get designs created with AI help. During some PR discussions the AI also indicated some problems it could not reproduce because they required dragging or similar direct interactions with the GUI. This implementation should allow those kinds of interactions. Perhaps the connection should allow exporting screenshots so that vision capable AIs could "look" at the same thing the user sees. It should also be able to tell which entities the user has selected to facilitate talking about "this" part of the sketch. The AI should be able to select sketch entities (by handle?) and apply constraints, do extrusions, switch groups, rotate the view to show the user particular geometry, and otherwise interact like a human designer working under an engineer for example.

Do not create any code for this yet. Claude, how feasible is this given what you know about solvespace internals? What approaches would you consider for doing this and why? What limitations would there be?

@ruevs and @BoykoNeov can you add anything (comments, ideas, etc) to this request and then turn Claude loose on coming up with a proposal? Or just tell me I'm being silly ;-)