Register Shiny Inputs and Outputs for MCP Access

Register shiny inputs and outputs for MCP (Model Context Protocol) agent access.

register_input() wraps a shiny input constructor to register metadata. It evaluates expr and returns the UI element, with tooltip as its hover tip (the HTML title attribute). as_tooltip() gives the default hover tip: the first sentence of description.

register_output() is a server-side function that registers a render function call (e.g. renderPlot({...})), assigns it to session$output, registers the MCP output spec, and sets up download-widget handlers. The UI overlay icons are injected entirely by JS.

Usage

register_input(
  expr,
  inputId,
  update,
  description = "",
  tooltip = as_tooltip(description),
  writable = TRUE,
  quoted = FALSE,
  env = parent.frame()
)

as_tooltip(description)

register_output(
  expr,
  outputId,
  description = "",
  quoted = FALSE,
  env = parent.frame(),
  ...,
  output_opts = list(),
  download_function = NULL,
  download_type = c("image", "htmlwidget", "threeBrain", "no-download", "data",
    "stream_viz"),
  extension = NULL,
  session = shiny::getDefaultReactiveDomain()
)

Arguments

expr

For register_input: a call expression that creates a shiny input widget. For register_output: a render function call such as renderPlot({...}).

inputId

character string. The shiny input ID (without the module namespace prefix).

update

character string. The fully qualified update function, e.g. "shiny::updateTextInput". Field mappings such as "shiny::updateSelectInput(value=selected)" override the default argument names passed to the update function.

description

character string. A human-readable description of the input or output purpose, exposed to LLM agents via MCP tools.

tooltip

character string shown when the mouse hovers over the input; defaults to the first sentence of description (see as_tooltip()). Set to NULL for no hover tip. The hover tip is added only when the outer tag of the UI element is an input (a button, a, input, select, or textarea tag, or a tag with class shiny-input-container) without a title; other elements, such as cards, are returned unchanged.

writable

logical (default TRUE). Whether the MCP update tool is allowed to change this input.

quoted

logical (default FALSE). If TRUE, expr is treated as already quoted; otherwise it is captured with substitute().

env

the environment in which to evaluate expr.

outputId

character string. The shiny output ID (without the module namespace prefix).

...

reserved for future use.

output_opts

a named list of extra options for the output (e.g. width, height defaults).

download_function

a custom download handler function. When download_type = "data", this function receives the file path and writes the download content.

download_type

character string. One of "image", "threeBrain", "data", or "no-download".

extension

character vector of allowed file extension for download, or NULL.

session

the shiny session object. For register_output, defaults to shiny::getDefaultReactiveDomain().

Value

register_input returns the evaluated UI element, with tooltip as the title of an input (see tooltip). as_tooltip returns the first sentence of description as a character string ("" when it is blank). register_output is called for its side effects (assigning the render function and registering widgets) and returns NULL invisibly.

See also

init_app, mcp_wrapper

Examples

# The default hover tip: the first sentence of the description
as_tooltip("Plot threshold, e.g. 0.5. Agents set it before running.")
#> [1] "Plot threshold, e.g. 0.5."

if (FALSE) { # \dontrun{
# inside a shidashi module UI function:
ns <- shiny::NS("demo")

register_input(
  expr = shiny::sliderInput(
    inputId = ns("threshold"),
    label = "Threshold",
    min = 0, max = 1, value = 0.5
  ),
  inputId = "threshold",
  update = "shiny::updateSliderInput",
  description = "Filter threshold for the plot"
)

# inside a shidashi module server function:
register_output(
  expr = renderPlot({ plot(iris) }),
  outputId = "my_plot",
  description = "Scatter plot of iris data",
  download_type = "image"
)
} # }