interactive-ui

Interactive UI

Two ways to put something in front of the user: a set of buttons to choose from, or a rendered panel built from Prefab components.

Prerequisites

• A client that renders MCP UI. Text-only clients will not show these
• No GitHub token, since neither tool calls GitHub

Choosing Between Them

choose when the answer is one of a few known options and you need the user to pick. github_pr_issue_analyser_ui when you have data worth showing as something other than a wall of text.

Neither replaces a plain answer. A single fact belongs in the reply.

choose

Parameter	Type	Default	Description
prompt	str	-	The question put to the user
options	list[str]	-	The options to show as buttons
title	str | None	None	Heading for the card, falls back to the provider default

The user clicks one option and the selection comes back into the conversation as a message. Nothing is returned to you directly, so treat the click as the user's next turn and carry on from it.

This is the right way to get confirmation before anything irreversible. The merge step in the pr-management skill requires an explicit yes from the user, and merge_pr does not prompt on its own, so ask here first:

choose(
  prompt="Merge PR #123 into main with a squash commit?",
  options=["Merge it", "Not yet"],
  title="Confirm merge",
)

It also suits picking a target when a search returned several, such as which of four open PRs to review, or which release version to tag.

Keep the options short enough to read on a button and make them distinct. Do not offer an option you are not prepared to act on.

github_pr_issue_analyser_ui

Parameter	Type	Default	Description
code	str	-	Prefab Python source to execute
data	str | dict | None	None	Initial data, as a dict or a JSON string

The code runs in a Pyodide WASM sandbox and renders as a streaming UI. Import everything you use, and make PrefabApp the outermost context manager so the panel renders progressively as the code is written: