Mcp Notifications

io.github.topvisor由 GitHub 帳號 topvisor 發布v1.0.3更新於 Oct 24, 2026

MCP server for desktop notifications over stdio

已驗證STDIO僅桌面效率與工作流開發者工具

概覽

AI 產生的概覽

一個本機 stdio MCP 伺服器,讓 AI 代理在任務完成、失敗或需要你輸入時發送系統桌面通知。

可以做什麼
  • 發送含標題與內容的桌面通知,可選擇提示音、自訂圖示與 Windows app_id 來源標籤。
  • 通知在背景佇列中送出,代理很快取得「Notification queued」結果而不會被阻塞。
  • 透過環境變數選擇投遞後端:自動、PowerShell 通知或 node-notifier 套件。
  • 可手動在對話中要求代理通知,也可在對話、專案或技能中加入指令以自動通知。
適用情境
適合在控制台類 AI 用戶端執行耗時或多步驟任務、而用戶端本身沒有通知功能時使用,避免一直查看聊天視窗。主要用於提示任務完成、失敗或需要你介入。
執行需求
以 stdio 本機程序執行,需從 npm 全域安裝 @topvisor/mcp-notifications,需要 Node.js 20 以上、npm 10 以上。需在 MCP 用戶端設定(例如 Codex 或 Claude)加入執行檔 mcp-notifications 並重新啟動用戶端。選用環境變數 MCP_NOTIFICATIONS_BACKEND 與 MCP_NOTIFICATIONS_APP_ID 可調整後端與 Windows 通知來源。
安裝前請注意
通知顯示在執行伺服器的本機上,代理寫入標題或內容的文字(包含任務細節或摘要)都會出現在桌面。自動通知指令若門檻設得太寬鬆,可能造成頻繁彈窗與提示音。選用圖示路徑會從本機檔案系統讀取。未宣告任何憑證或網路存取。

安裝

在 SourceWeft 中

  1. 開啟 儀表板中的 Mcp Notifications,將其新增到工作區。
  2. 為需要使用其工具的對話啟用該服務。

Desktop only,透過 STDIO。 STDIO 服務會啟動本機處理程序,因此需要 SourceWeft 桌面主機。

其他 MCP 客戶端

參照 儲存庫 中的啟動說明。

README

mcp-notifications

[npm version]

🚀 Ship faster. Miss nothing.

mcp-notifications lets your MCP agent send desktop notifications for completed tasks, failures, and important updates.

This is especially useful for console AI agents, because they usually have no built-in notifications by default.

No more checking chats every minute.
No more silent agent responses.
Just visible, reliable feedback on your desktop. 🔔

Common problem this solves

You give tasks to several AI agents, then wait and keep checking who already replied.

With mcp-notifications, this gets simpler: the agent can notify you on desktop when it needs your input or when work is done.

Why teams install this

  • ⚡ Immediate feedback from your MCP workflows.
  • 🧠 Better focus: let the agent work while you stay on your main task.
  • 🖥️ Native OS notifications via node-notifier.
  • 🧵 Non-blocking behavior: notification sending is queued in background.

Install globally

Requirements:

  • Node.js >=20
  • npm >=10
bash
npm i -g @topvisor/mcp-notifications

Executable name (use this in MCP config):

bash
mcp-notifications

mcp-notifications is an MCP server entrypoint (stdio), not a one-shot notification command.

Setup: Codex

Add this to ~/.codex/config.toml:

toml
[mcp_servers.notifications]enabled = truecommand = "mcp-notifications"args = []

Restart Codex after config update.

Setup: Claude Agent

Add MCP server config to your Claude client config file:

json
{  "mcpServers": {    "notifications": {      "command": "mcp-notifications",      "args": []    }  }}

Then restart Claude client.

AI Agent Notification Instructions

There are two approaches to notifications: manual and automatic.

Choose the one that works better for your workflow.

Manual

In a task where you want to be notified, explicitly ask the agent. Example:

text
Count files in the project; after the task is fully complete, notify me with sound.

You can also define sound, topic, and frequency rules inside a specific chat. Example:

text
Notify me about your replies without sound, include the reply text, and use title: "Large Refactoring"

Automatic

Automatic notifications can be configured globally or per project.

Example instruction to enable automatic notifications for agent replies:

text
Send `send_notification` after your replies (actual task time >5 seconds or many steps); `play_sound: false`; `app_id: '{put your chat name here}'`

Time is a rough threshold and depends on model behavior, so adjust this instruction to your own preferences.

In Skills

You can also enable notifications in specific skills. Example for a Review skill:

text
After the review, run `send_notification` with a short summary; `play_sound: true`; `app_id: 'Reviewer {put task id here}'`

Tool

send_notification

Input:

  • title string (required)
  • message string (required)
  • play_sound boolean (optional, default: false)
  • icon string (optional, absolute or relative path to image file)
  • app_id string (optional, Windows App User Model ID for toast source)

Example:

json
{  "title": "Codex",  "message": "Deployment completed successfully",  "play_sound": true,  "icon": "/opt/mcp-notifications/icons/custom.png",  "app_id": "Topvisor.Codex"}

app_id (Windows)

  • app_id controls the source shown in Windows toast notifications.
  • If app_id is not set, Windows may show SnoreToast as the source.
  • You can pass app_id in each tool call, or set MCP_NOTIFICATIONS_APP_ID as an environment variable for the server.

Backend selection

You can select notification transport via env MCP_NOTIFICATIONS_BACKEND:

  • auto (default): uses powershell in WSL if available, otherwise node-notifier
  • powershell: force powershell.exe Windows toast transport (recommended for WSL)
  • node-notifier: force node-notifier package

Chat Prompts To Test In Codex

Use these messages directly in chat:

text
Check send_notification
text
Send a notification without sound: title "Test", message "Check"
text
Send a notification with sound: title "Test", message "Check"
text
Send a notification with app_id "Topvisor.Codex": title "Test", message "Check"
text
Send 3 test notifications in a row without sound

Expected tool result in logs/response:

text
Notification queued

Behavior

  • ✅ Uses standard system notification channels.
  • 🔊 Uses the standard system notification sound when play_sound: true.
  • 🤖 Uses bundled Topvisor robot image as default notification icon.
  • 🧰 Returns quickly while notifications are delivered in background queue.

Example

The AI agent is waiting for you:

[example.png]

來源:README.md,提交 9cd5ee5

工具

0
SourceWeft 尚未在沙箱中啟動過該軟體包,因此其工具未知。

版本歷史

1
  1. v1.0.3最新Sep 16, 2026

評分

已滿足適用檢查項的 50%

該評分衡量此收錄條目的完整程度,依據註冊中心條目、端點與公開來源計算得出。它不是對伺服器本身的評價,也不表示伺服器安全或目前可用:可連通性是帶日期的檢查結果,評分也不包含安裝量或使用者評價。

  • 近 30 天內可連通25 分

    請保持端點能夠回應 MCP 握手;SourceWeft 每 30 天探測一次。

  • 已宣告認證方式10 分

    請在註冊中心條目中宣告憑證請求標頭,或保持端點能夠回應探測,以便判定其認證方式。

  • 已列出工具15 分

    請讓伺服器在無憑證時回應 tools/list,或發布可在沙箱中無憑證啟動的軟體包。

  • 描述必需15 分
  • 圖示10 分
  • 文件15 分
  • 授權條款10 分