Skip to contents

mv_mcp_server() starts a Model Context Protocol server that lets an AI assistant (Claude Desktop, Claude Code, OpenAI Codex, ChatGPT, the OpenAI Agents SDK, or any other MCP client) work on your mirtverse teaching materials. You describe what you want ("add a question on local dependence to Lab 1", "use my own exam data in the fairness lab", "turn my ltm lecture notes into mirtverse") and the assistant does it with the server's tools: It reads and edits the files in your course folder, copies and builds labs with the instructor toolkit, looks up mirtverse functions so it does not invent any, translates older ltm, eRm, TAM and irtQ code with mirtify(), and runs the materials to check that they still work.

Usage

mv_mcp_server(
  course = ".",
  transport = c("stdio", "http"),
  allow_run = NULL,
  host = "127.0.0.1",
  port = 8765,
  token = NULL,
  timeout = 600
)

Arguments

course

path to the course folder the assistant may work in. It is created if it does not exist. Defaults to the working directory.

transport

"stdio" (the default, for desktop and command-line clients that start the server themselves) or "http" (a Streamable HTTP endpoint, for clients such as ChatGPT that connect to a URL).

allow_run

logical; may the assistant run R code (the check_material tool and running translated code)? Defaults to TRUE for "stdio" and FALSE for "http".

host, port

where the HTTP server listens.

token

for "http", a secret that becomes part of the URL (http://host:port/mcp/<token>), so only people who know the URL can reach the server. A random token is made if NULL; use "" for none.

timeout

seconds a check may run before it is stopped.

Value

Runs until the client disconnects (stdio) or the R session is interrupted (http); returns NULL invisibly.

Details

The server can only read and write files inside course. Every change is backed up first (in course/.mirtverse-mcp/backups) and can be undone with the undo_last_change tool.

You normally do not call mv_mcp_server() yourself. Your AI client starts it; mv_mcp_config() prints the settings to paste into the client.

See also

mv_mcp_config() for client setup, and the "AI assistant for instructors" article.

Examples

if (FALSE) { # \dontrun{
# Usually started by the AI client, with a command such as
#   Rscript -e "mirtverse::mv_mcp_server('~/courses/edpy507')"
mv_mcp_server("~/courses/edpy507")

# For ChatGPT: serve over HTTP, then expose the port with a tunnel
mv_mcp_server("~/courses/edpy507", transport = "http", port = 8765)
} # }