[Feature] Give us an API hook for MCP servers to switch the session's AI model

[iandouglas]

I'm working on an idea to create an MCP server that allows users to switch to a preferred LLM based on the prompt they type. This would require an API hook of some kind into Void to change the LLM for that session. (this might get complicated if I have several Void editors open?)

If I were to write a prompt like help me write some Python code to ... then that API hook would allow the MCP to change that session's LLM to my preferred coding LLM.

Later if I say "write me a blog post about this project," the MCP server could switch to my preferred LLM for writing.

There's obviously some complexity to work out for when to keep using the previous LLM choice or when to switch, or when to use a "fallback" LLM if nothing else matches, but you get the idea :)

{
  "coding": {
    "keywords": ["software", "development"],
    "provider": "Anthropic",
    "model": "claude-sonnet-4"
  },
  "writing": {
    "keywords": ["story", "blog", "poem"],
    "provider": "OpenAI",
    "model": "gpt-4.1"
  },
  "**FALLBACK**": {
    "provider": "Google",
    "model": "gemini-flash-lite"
  }
}

[andrewpareles]

Interesting idea! If we let MCP servers change the current model, we might as well also have them change the parameters like how much thinking to do and what tools to use (although letting an MCP server switch to 'agent' mode is a little dangerous). Will think about this - just don't want to add too much complexity/upkeep for ourselves for a non essential feature.