mcpField
M
mcp-notifications

mcp-notifications

Local
github.com/ArtemeeyProductivity2Updated 24w ago

About mcp-notifications

mcp-notifications is an MCP server that lets console agents send desktop notifications when tasks finish, fail, or need input. It queues delivery in the background and supports sound and notification metadata.

Key features

  1. 01Send desktop notifications
  2. 02Set title and message
  3. 03Enable notification sound
  4. 04Set icon and app ID
  5. 05Queue delivery in the background

Use cases

  1. 01Get notified when an agent finishes a task
  2. 02Notice failures or requests for input
  3. 03Receive updates while focused on other work

MCP support

The stdio MCP server exposes the `send_notification` tool to MCP clients, with options for sound, icon, and app ID.

Configuration

Local server config is the same for every client. Copy it into your client’s MCP config file.

mcp.json · Universal config
{
"mcpServers": {
"topvisor-mcp-notifications": {
"command": "npx",
"args": [
"-y",
"@topvisor/mcp-notifications"
]
}
}
}
Run commandnpx -y @topvisor/mcp-notifications

Local servers need a runtime on your machine (e.g. Node.js 18+). Config file locations differ by client; see your client’s docs.

Packages
npm@topvisor/mcp-notifications1.0.3

Tools

1

Saved tool list · updated 2w ago

ParameterTypeRequiredDescription
iconstringNo
titlestringYes
app_idstringNo
messagestringYes
play_soundbooleanNo

Details

From the project 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
npm i -g @topvisor/mcp-notifications

Executable name (use this in MCP config):

mcp-notifications

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

Setup: Codex

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

[mcp_servers.notifications]
enabled = true
command = "mcp-notifications"
args = []

Restart Codex after config update.

Setup: Claude Agent

Add MCP server config to your Claude client config file:

{
  "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:

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:

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:

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:

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:

{
  "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:

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

Expected tool result in logs/response:

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

Comments

Markdown supported

New to MCP?

From what the protocol is to installing your first server, these answers cover most first questions.

An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.