ask-user-questions
Enables AI models to ask users questions through a local web interface, supporting batch questions, multi-select, and free text for human-in-the-loop interactions.
README
ask-user-questions
An MCP (Model Context Protocol) tool that allows AI models to ask users questions via a local web interface, enabling human-in-the-loop interactions.
Features
- Daemon architecture: Independent background process for HTTP/WS, supports multiple MCP clients simultaneously
- Batch questions: Up to 4 sub-questions per call, with single/multi-select and free text
- Rich options: Options support label, description, and recommended flags
- Free text: Every sub-question always includes an "Other" free text input
- Routed navigation: Home page with pending questions + history; detail page for each question
- Internationalization: 5 languages (English, 中文, 한국어, 日本語, Русский)
- Dark mode: System / Light / Dark themes
- Responsive: Desktop and mobile friendly
- Configurable: Timeout, notification toggle, auto-open browser toggle
Quick Start
1. Configure MCP Client
ask-user-questions provides an MCP STDIO server as a tool interface, which can connect to any MCP client you prefer, such as Claude Desktop or VS Code.
Claude Desktop (claude_desktop_config.json):
{
"mcpServers": {
"ask-user-questions": {
"command": "npx",
"args": ["ask-user-questions"]
}
}
}
2. Get Started
- When the model invokes the
ask_usertool during a conversation, the browser interface will automatically open to display the question. - Use the browser interface to view pending questions, history, and details for each question, then submit your answers.
- The model receives your responses in real time, enabling seamless human-in-the-loop interaction.
MCP Tool
ask_user
Ask the user questions and wait for their response. Supports batch questions (1–4 sub-questions), each with configurable options and multi-select mode. An "Other" free text input is always available—no need to add it manually.
Parameters:
| Parameter | Type | Required | Description |
|---|---|---|---|
questions |
SubQuestion[] |
✅ | Array of 1–4 sub-questions |
SubQuestion:
| Field | Type | Required | Description |
|---|---|---|---|
question |
string |
✅ | Question text |
multiSelect |
boolean |
❌ | Allow multiple selections (default false) |
options |
Option[] |
❌ | Preset options (omit for free text only) |
Option:
| Field | Type | Required | Description |
|---|---|---|---|
label |
string |
✅ | Option label |
description |
string |
❌ | Option description |
recommended |
boolean |
❌ | Mark as recommended |
Example:
{
"name": "ask_user",
"arguments": {
"questions": [
{
"question": "Which API style do you prefer?",
"options": [
{ "label": "Composition API", "recommended": true },
{ "label": "Options API" }
]
},
{
"question": "Which features do you need?",
"multiSelect": true,
"options": [
{ "label": "TypeScript", "description": "Add type support" },
{ "label": "i18n", "description": "Internationalization" },
{ "label": "Dark Mode" }
]
}
]
}
}
Configuration
Config file: ~/.ask-user-questions/config.json (also editable via the Web UI settings panel).
{
"timeout": 0,
"notification": true,
"autoOpenBrowser": true
}
| Option | Type | Default | Description |
|---|---|---|---|
timeout |
number |
0 |
Timeout for waiting (ms), 0 = no timeout |
notification |
boolean |
true |
Show desktop notifications |
autoOpenBrowser |
boolean |
true |
Auto-open browser |
How It Works
┌──────────┐ stdio ┌──────────┐ HTTP ┌────────────────┐ WebSocket ┌──────────┐
│ AI Model │◄───────►│ MCP STDIO│────────►│ Daemon Server │◄───────────►│ Web App │
│ (Claude, │ │ (bin.mjs)│ proxy │ (daemon.mjs) │ │ (Browser)│
│ GPT...) │ └──────────┘ │ localhost:13390│ └──────────┘
└──────────┘ └────────────────┘
▲ HTTP proxy ▲
┌──────────┐ stdio ┌──────────┐ │ │
│ AI Model │◄───────►│ MCP STDIO│─────────┘ │
│ (another) │ │ (bin.mjs)│ │
└──────────┘ └──────────┘ ┌───────┴──────┐
│ Store │
│ (in-memory) │
└──────────────┘
- First MCP launch:
bin.mjsauto-spawns a background daemondaemon.mjs(HTTP + WebSocket server on port 13390) - Subsequent MCP launches: Detects the daemon is already running and reuses it
- All MCP instances proxy to the daemon's Store via HTTP API (create questions + long-poll for answers)
- The daemon pushes real-time events to the browser via WebSocket
- The daemon detects browser presence via WebSocket connection count—only opens the browser when no clients are connected
Development
Prerequisites
- Node.js >= 20
- pnpm >= 9
Local Development
# Install dependencies
pnpm install
# Start App dev server (with HMR)
pnpm dev
# Start Server dev server (HTTP only)
pnpm dev:server
# Build all packages
pnpm build
# Type checking
pnpm typecheck
Project Structure
ask-user-questions/
├── packages/
│ ├── server/ # Daemon + MCP Shell + HTTP Server
│ │ └── src/
│ │ ├── bin.ts # MCP entry: ensure daemon + start MCP STDIO
│ │ ├── daemon.ts # Daemon entry: background HTTP+WS server
│ │ ├── server.ts # Hono + @hono/node-ws HTTP/WS server
│ │ ├── mcp.ts # MCP stdio + tool definitions (HTTP API proxy)
│ │ ├── store.ts # In-memory state + event pub/sub
│ │ ├── config.ts # Config file management
│ │ ├── notify.ts # Desktop notifications + browser management
│ │ ├── types.ts # Type definitions
│ │ └── index.ts # Public API exports
│ └── app/ # Vue 3 Web UI
│ └── src/
│ ├── App.vue # Shell layout (header + router-view)
│ ├── router.ts # Vue Router config
│ ├── pages/ # Page components
│ │ ├── home.vue # Home: pending list + history
│ │ └── question-detail.vue # Question detail page
│ ├── components/ # UI components
│ │ ├── question-card.vue # Question card (batch sub-questions)
│ │ └── settings-panel.vue # Settings panel
│ ├── composables/ # Vue composables
│ │ ├── use-questions.ts # Question state + WebSocket
│ │ └── use-dark-mode.ts # Dark mode
│ └── lib/
│ ├── api.ts # API client + WebSocket
│ └── i18n.ts # i18n (5 languages)
├── docs/ # Architecture docs
├── .github/workflows # CI/CD
└── pnpm-workspace.yaml
See Architecture (English) | 架构文档 (中文) for details.
Publishing
This project uses GitHub Actions + npm provenance (OIDC):
# 1. Bump version
cd packages/server
npm version patch # or minor / major
# 2. Push tag
git push --follow-tags
# 3. GitHub Actions auto-builds and publishes to npm
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。