Smart Appliance MCP
Smart Appliance MCP
一个 MCP 服务器,让任何支持 MCP 的 LLM 客户端都能发现并控制本地网络上的智能家电。
重要的设计选择是适配器驱动的路由:
discover_devices使用每个已注册的适配器进行扫描。每个已发现的设备都会存储其
provider。后续调用仅使用
deviceId;服务器查找设备并将命令路由到发现期间找到的适配器。
这使客户端提示保持简单。LLM 客户端无需知道电视是 Roku、Home Assistant、Samsung、LG、Matter 还是其他品牌。
工具
discover_devices:查找家电并在此服务器会话中记住它们。diagnose_discovery:解释发现状态和可能的网络阻塞因素,无需用户编辑技术配置。list_known_devices:返回已找到的设备。get_device_controls:显示单个设备的可用能力。discover_apps:探测已发现媒体设备暴露的应用启动目标。search_apps:按名称、包 ID、类别提示和可启动性搜索已发现的应用。pair_device:当提供商要求时启动一次性消费者配对。complete_pairing:使用设备上显示的代码完成配对。list_pairings:列出本地存储的配对。remove_pairing:移除本地存储的配对。control_device:执行音量、导航、电源、搜索和应用启动等操作。search_content:在支持时搜索已安装应用或原生内容提供商。suggest_content:返回适配器感知的观看建议。record_watch_event:记住已观看、喜欢、忽略或开始的内容。list_watch_history:显示推荐使用的近期本地观看历史。recommend_content:按类别、新鲜度、应用、观看历史和应用可启动性对接下来要观看的内容进行排序。get_device_state:当适配器支持时返回状态。
Related MCP server: MCP Roku Control
包含的适配器
roku:通过 SSDP 发现 Roku 电视和 Roku 流媒体设备,并通过 Roku ECP 控制它们。smart_appliance_companion:通过 mDNS 发现可选的电视端配套应用,并使用它进行已安装应用列表和包启动。google_tv_remote:通过 mDNS Google Cast 信号和 DIAL/SSDP 发现 Google TV / Android TV 设备,然后模拟正常的遥控器式配对流程。home_assistant:可选的广泛设备桥接,适用于电视、灯光、开关、恒温器等。google_tv:仅用于开发/测试的可选 ADB 回退。使用ENABLE_ADB_ADAPTER=true启用。
快速开始
npm install
npm run build
npm start用于本地开发:
npm run dev客户端配置
构建项目,然后将如下服务器条目添加到你的 MCP 客户端:
{
"mcpServers": {
"smart-appliance": {
"command": "node",
"args": ["/absolute/path/to/smart-appliance-mcp/dist/index.js"]
}
}
}如果你使用 Home Assistant,请包含:
{
"env": {
"HOME_ASSISTANT_URL": "http://homeassistant.local:8123",
"HOME_ASSISTANT_TOKEN": "your-long-lived-access-token"
}
}对于 Google TV / Android TV,使用消费者配对流程。服务器通过局域网信号(如 mDNS _googlecast._tcp.local 和 DIAL/SSDP)发现电视,然后在已发现设备记录上内部携带地址。
预期的用户流程是:
Discover my smart appliances.
Pair my living room TV.
Complete pairing with code 123456.
Turn the TV volume up.已发现的 Google TV 设备使用 provider: "google_tv_remote"。配对状态存储在本地,并通过与所有其他提供商相同的适配器注册表进行路由。
Google TV 远程适配器包括本地发现、消费者配对、实时远程控制、应用启动探测和适配器路由的命令执行。ADB 适配器和 GOOGLE_TV_REMOTE_DEVICES 覆盖仅作为可选的开发诊断工具保留,而非正常用户设置。
应用发现也刻意采用适配器驱动。在 Google TV 上,discover_apps 探测电视本地暴露的启动表面,例如 DIAL /apps/<name> 端点。如果电视未通过消费者远程或 DIAL 接口暴露已安装应用列表,服务器会明确报告,而不是假装猜测的包名或浏览器 URL 是已发现的应用启动路径。
为了获得最佳 Google TV 体验,请从 companion/google-tv 安装可选的配套应用。该配套应用在电视上运行,广播 _smart-appliance._tcp.local,使用 Android PackageManager 列出已安装的 Leanback 启动器应用,并按包名在本地启动应用。这是针对 Crunchyroll 等不暴露 DIAL 启动端点的应用的正常用户路径。
推荐
推荐层是本地优先且适配器感知的:
当已发现设备可以提供时,优先使用适配器观看历史。
record_watch_event存储轻量级本地回退历史,包括应用、标题、类别、进度和状态。recommend_content合并电视来源的历史、本地回退历史、提供的内容目录和起始行。结果按新鲜度、与近期观看的类别重叠、应用可用性、可启动性以及已看/已忽略状态进行评分。
响应将新鲜的
recommendations与alreadyWatched和dismissed匹配分开。每条推荐都包含
userSummary/userReasons以提供清晰面向用户的答案,以及用于内部规划的详细字段。可操作的推荐行包含插图和操作:
artwork.thumbnailUrl、posterUrl和backdropUrl用于图片。当目录提供
previewUrl/trailerUrl时,actions.preview用于预告片或预览片段。actions.primary作为一键观看操作,表示为 MCP 工具调用负载。
chatCards和format_recommendation_cards为聊天客户端渲染相同的结果:仅在目录提供标题特定图片时才包含图片。
预览链接使用正常 Web URL。
观看/搜索链接使用
mcp://action?...URL,描述 MCP 工具调用供主机客户端确认和执行。
如果当前适配器无法提供电视观看历史,list_watch_history、recommend_content 和 format_recommendation_cards 会返回可选的 companionPrompt。Google TV 的消费者远程协议不暴露私有的按应用流媒体历史,因此精确的内容历史需要提供商集成或可选的电视端配套来源。
提供商目录不断变化,因此生产客户端应从提供商集成、搜索连接器或用户拥有的媒体源将新鲜目录行传入 recommend_content。除非适配器或连接器提供,否则 MCP 不声称拥有实时的 Netflix/Crunchyroll 目录。
前端观看队列
运行本地 UI 以获取可操作的推荐:
npm run ui打开 http://localhost:5177。UI 将 MCP 推荐输出渲染为卡片,包含插图、预览、一键观看操作、筛选、搜索和已观看面板。使用 {} 按钮粘贴来自任何 LLM 客户端的 recommend_content 响应。
对于聊天原生卡片,使用与 recommend_content 相同的输入调用 format_recommendation_cards。它返回 cards 以及 Markdown,使用链接而不是按钮。默认情况下,链接指向本地 UI 操作端点,因此请保持 npm run ui 运行:
[Watch on TV](http://127.0.0.1:5177/api/actions/run?payload=...)如果主机客户端直接支持 mcp://action?... 链接,请使用 linkMode: "mcp_scheme"。
示例工具流程
首先询问客户端:
Discover my smart appliances.然后:
Turn the living room TV volume up.MCP 服务器在内部处理路由:
const device = registry.getDevice(deviceId);
const adapter = registry.adapterFor(device);
await adapter.control(device, request);添加新适配器
创建一个实现 SmartApplianceAdapter 的类:
export class SamsungTizenAdapter implements SmartApplianceAdapter {
readonly id = "samsung_tizen";
readonly label = "Samsung Tizen TV";
async discover(options: DiscoveryOptions): Promise<SmartDevice[]> {
return [];
}
async control(device: SmartDevice, request: ControlRequest) {
return { ok: true };
}
}然后在 src/index.ts 中注册它:
registry.register(new SamsungTizenAdapter());发现始终是事实来源。一旦使用 provider: "samsung_tizen" 发现三星电视,该 deviceId 的所有后续命令都会自动路由到三星适配器。
备注
本地网络发现取决于你的网络是否允许多播/SSDP。
某些电视生态系统在控制前需要配对;这些适配器应将配对流程作为 MCP 工具或资源暴露。
内容推荐是适配器感知的,但可以通过结合设备能力与主机 LLM 客户端的品味/偏好上下文来增强。
Available Tools
18 toolscomplete_pairingA
Complete consumer pairing for a discovered device using the code shown by the appliance.
Inputs: code (required), deviceId (required).
control_deviceD
Control a discovered device. The server routes the command through the adapter found during discovery.
Inputs: action (required), deviceId (required), value, payload.
diagnose_discoveryA
Explain current discovery state and next checks without asking users for technical setup.
Inputs: no named parameters.
discover_appsC
Probe app launch targets exposed by a discovered media appliance.
Inputs: deviceId (required).
discover_devicesB
Discover smart appliances on the local network and remember their provider adapter.
Inputs: timeoutMs.
format_recommendation_cardsC
Format recommendations as chat-renderable cards with preview links and MCP action links instead of buttons.
Inputs: deviceId (required), apps, mood, limit, +7 more.
get_device_controlsB
Return available controls for a discovered device.
Inputs: deviceId (required).
get_device_stateA
Return current state for a discovered device when the discovered adapter supports it.
Inputs: deviceId (required).
list_known_devicesA
List devices discovered during this MCP server session.
Inputs: no named parameters.
list_pairingsA
List locally stored consumer pairings.
Inputs: no named parameters.
list_watch_historyC
List recent watch history from the TV adapter when available, merged with local MCP history.
Inputs: limit, deviceId.
pair_deviceA
Start consumer pairing for a discovered device when its provider supports pairing.
Inputs: deviceId (required).
recommend_contentB
Rank what to watch next using watch history, categories, freshness, apps, and discovered launchability.
Inputs: deviceId (required), apps, mood, limit, +7 more.
record_watch_eventA
Record watch activity so recommendations can avoid repeats and learn category/app preferences.
Inputs: title (required), appId, source, status, +4 more.
remove_pairingB
Remove a locally stored consumer pairing.
Inputs: deviceId (required).
search_appsA
Search all apps exposed by a discovered media appliance, including app names, package ids, categories, and launchability.
Inputs: deviceId (required), limit, query, onlyLaunchable.
search_contentB
Search content or installed apps on a discovered media appliance using its discovered adapter.
Inputs: query (required), deviceId (required).
suggest_contentB
Suggest content to watch based on a discovered media appliance and optional viewing prompt.
Inputs: deviceId (required), prompt.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 18 tools
Most tools target a distinct action/resource (discovery, pairing, device control, media recommendations). Minor overlap exists between suggest_content and recommend_content, and between search_content and search_apps, but descriptions clarify different inputs and purposes.
All tools use a consistent snake_case verb-first naming convention (list_, discover_, get_, control_, pair_, search_, suggest_, etc.). No mixed styles or arbitrary names; resource nouns follow verbs predictably.
At 18 tools the surface is on the heavy side for a single server, covering several subdomains (discovery, pairing, media, device control). The tools are not redundant enough to cut, but the count is above the ideal 3-15 range.
The server covers the full pairing lifecycle and core discovery/control/media workflow with no dead ends. Minor gaps like explicit device forgetting or media playback commands are absent, but adapters likely make those device-specific.
Maintenance
Related MCP Connectors
Control Android TV from any AI. 38 MCP tools: playback, recap, recommend, smart-home, schedules.
Control your existing Philips lights, Sonos speakers, Roborock vacuums and other supported devices through Port OS cloud. Give ChatGPT, Claude and other MCP-compatible assistants access to the devices you connect. Turn off the living room lights, play music on your speakers, adjust the thermostat, run a vacuum in a room, check a camera snapshot or find a connected tracker. Available actions depend on your linked provider accounts and each device’s supported capabilities. Port OS runs in the cloud; no local server is required. You choose the account, device and capability permissions each assistant receives. Device operations require user OAuth; public tool discovery does not grant device access. Port OS is in alpha. Setup: https://portoperatingsystem.com/docs/getting-started/ Support: https://portoperatingsystem.com/contact/
Give your AI agent the power to cast to your home TV and control an interactive canvas. By ArdaBot.
- DNS-DATA RemoteDeprecatedcom.dns-data
Connect AI assistants to authorized Windows and macOS devices for files, commands, and management.
Related MCP Servers
- AlicenseBqualityAmaintenanceControl any smart TV with natural language — play Netflix/YouTube/Spotify by name, cast URLs, scene presets, multi-room audio, and multi-TV sync. Supports LG, Samsung, Android TV, Roku.2373 PyPI60MIT
- FlicenseAqualityDmaintenanceEnables AI assistants to control Roku TVs on the local network via natural language commands.6-
- AlicenseBqualityDmaintenanceEnables AI assistants to discover and control IoT devices on local networks and via Bluetooth LE.71MIT
- AlicenseNot gradedqualityBmaintenanceEnables AI assistants to discover and control smart TVs on the local network, including power, volume, navigation, and app launching across multiple brands.MIT