Clarify user-facing role (precedence?) of btw vs mcptools #116

Hi again folks.

Apologies for the spate of issues, but I'm prepping an internal workshop on R-based agentic tools... and, in so doing, encountering issues that I'll need to have an informed answer for.

With the impending (hopefully!) submission of btw to CRAN, I'm wondering if you could offer guidance on which of mcptools or btw end-users should typically be using to interact with their R sessions?

My current understanding is that mcptools has moved towards being a low-level infra package and that btw now provides the main user-facing functionality. So, the basic workflow is (1) add the btw MCP server to your mcp.json

{
  "mcpServers": {
    "r-btw": {
      "command": "Rscript",
      "args": ["-e", "btw::btw_mcp_server()"]
    }
  }
}

and then (2) enable access to your session with btw::btw_mcp_session().

That seems reasonable to me... BUT I've also seen official guidance that inverts this relationship. For example, this (admittedly slightly older) blog post suggests that we should provision the btw tools through mcptools::mcp_server():

{
  "mcpServers": {
    "r-mcptools": {
      "command": "Rscript",
      "args": ["-e", "mcptools::mcp_server(tools = btw::btw_tools())"]
    }
  }
}

before calling mcptools::mcp_session() to establish the session connection.

I can imagine the answer being that both approaches are valid and that it doesn't matter much in the end. But some clarity from the dev perspective about the precedence and ordering of these two packages would be appreciated.

Cheers and thanks again for working on these extremely helpful set of tools!

P.S. I guess another way of asking my question is: Do I lose anything (code injection?, etc.) by going through btw::btw_mcp_server() -> btw::btw_mcp_session(), rather than mcptools::mcp_session() -> mcptools::mcp_server(tools = btw::btw_tools())?