Wrap an MCP Tool Generator Function — mcp

Creates a wrapper around a generator function to ensure it returns valid Model Context Protocol (MCP) tool definitions. The wrapper validates input and filters output to contain only 'ellmer::ToolDef' objects.

Usage

mcp_wrapper(generator)

Arguments

generator: A function that accepts a `session` parameter and returns either a single tool object or a list/vector of such objects; see tool.

Value

A wrapped function with class 'shidashi_mcp_wrapper' that: - Accepts a `session` parameter - Calls the generator function with the session - Normalizes the output to a list - Filters to keep only valid tool objects - Returns a list of tool objects (possibly empty)

Details

The wrapper performs the following validations: - Ensures `generator` is a function - Checks that `generator` accepts a `session` parameter

The returned function automatically handles both single tool definitions and lists of tools, providing a consistent interface for MCP tool registration.

Examples

# Define a generator function that returns tool definitions
my_tool_generator <- function(session) {
  # Define MCP tools using ellmer package

  tool_rnorm <- tool(
    function(n, mean = 0, sd = 1) {
      shiny::updateNumericInput(session, "rnorm", value = rnorm)
    },
    description = "Draw numbers from a random normal distribution",
    arguments = list(
      n = type_integer("The number of observations. Must be positive"),
      mean = type_number("The mean value of the distribution."),
      sd = type_number("The standard deviation of the distribution.")
    )
  )

  # or `list(tool_rnorm)`
  tool_rnorm
}

# Wrap the generator
wrapped_generator <- mcp_wrapper(my_tool_generator)