MCPal
Friendly Companion MCP. Get notified when your task is finished!
Documentation
MCPal
Lightweight MCP server for native desktop notifications with action buttons, text replies, and LLM-aware icons.
Compatible with any MCP client.
Usage
Add to your MCP Config manually:
// JSON
{
"mcpServers": {
"mcpal": {
"command": "npx",
"args": ["mcpal@latest"]
}
}
}Or TOML (Codex):
// TOML
[mcp_servers.mcpal]
command = "npx"
args = ["-y", "mcpal@latest"]Codex CLI
Add MCPal directly via the CLI:
codex mcp add mcpal -- npx -y mcpal@latestClaude Code
Add MCPal directly via the CLI:
# Project scope (current project only)
claude mcp add mcpal -- npx -y mcpal@latest
# User scope (available in all projects)
claude mcp add mcpal --scope user -- npx -y mcpal@latestConfiguring Proactive Notifications
MCPal includes built-in MCP instructions that tell clients *how* to use notifications. To configure *when* your assistant should proactively notify you, add this to your client instruction file. This should either be `AGENTS.md` or `CLAUDE.md` based on your model:
## Post Task Completion Hook
ALWAYS use MCPal MCP tool (if available) to notify me when you finish any tasks (this includes reading files, investigating, coding). You should use this tool before you terminate conversations. Refer to tool instructions precisely to craft your response.Customize this based on your preferences - some users want frequent updates, others only want notifications for important decisions.
Other MCP Clients
MCPal works with any MCP-compatible client (Cursor, VS Code with MCP extensions, etc.). The setup varies by client - refer to your client's documentation for adding MCP servers.
Tool: `send_notification`
Send native notifications with optional features.
Parameters
| Parameter | Type | Required | Description |
|---|---|---|---|
| `message` | string | Yes | The notification body text |
| `title` | string | No | The notification title (default: "MCPal") |
| `actions` | string[] | No | Action buttons (e.g., `["Yes", "No", "Maybe"]`) |
| `dropdownLabel` | string | No | Label for actions dropdown (required for multiple actions) |
| `reply` | boolean | No | Enable text reply input |
Examples
Simple notification:
{
"message": "Build complete!",
"title": "CI/CD"
}With actions:
{
"message": "Deploy to production?",
"title": "Deployment",
"actions": ["Deploy", "Cancel"],
"dropdownLabel": "Choose"
}With reply:
{
"message": "What should I name this file?",
"title": "Question",
"reply": true
}You can reply directly from the notification without switching apps:
Tool Result Contract
`send_notification` now returns a dual contract:
- Canonical machine output via `structuredContent` (recommended for parsing)
- Backward-compatible text output in `content[0].text`
Structured fields:
- `status`: `"sent"` or `"error"`
- `title?`: Notification title
- `message?`: Message actually sent after sanitization
- `response?`: Notification response (`"timeout"`, clicked action, etc.)
- `activationType?`: Activation source (`"replied"`, `"actionClicked"`, etc.)
- `reply?`: User free-form reply
- `error?`: Error message when `status` is `"error"`
- `sanitized?`: `true` when MCPal had to sanitize or truncate inputs
Legacy text is still line-based, but each value is JSON-encoded on a single line for parser safety, for example:
status: "sent"
title: "MCPal"
message: "Line 1\nLine 2"
response: "timeout"Input Sanitization
Before delivery, MCPal applies best-effort sanitization to reduce notifier/parser failures:
- Normalize line endings: `\r\n` / `\r` -> `\n`
- Remove unsafe control chars (keeps `\n` and `\t`)
- Truncate limits:
- `title`: 256 chars
- `message`: 4000 chars
- `actions`: max 3 items, each 64 chars
- `dropdownLabel`: 64 chars
LLM-Aware Icons
MCPal detects which MCP client is calling the tool and displays the appropriate icon in notifications.
| Client | Icon |
|---|---|
| Claude Desktop / Claude Code / Opus | Claude logo |
| Codex / OpenAI / ChatGPT | OpenAI logo |
| Cursor | Cursor logo |
| VS Code | VS Code logo |
| Unknown | No icon |
This works via the MCP protocol's client identification - each client sends its name during initialization.
Adding New Client Icons
To add support for a new LLM client, add a PNG to `src/assets/clients/` and update the mapping in `src/notify.config.ts`.
Icon Specifications:
| Property | Requirement |
|---|---|
| Format | PNG with transparency (RGBA) |
| Dimensions | 128×128 pixels |
| File size | Notifications > MCPal** |
Development
# Install dependencies
pnpm install
# Build (required after clone - sets up desktop notification app)
pnpm run build
# Type check
pnpm run typecheck
# Lint
pnpm run lint:fix
# Format
pnpm run format:fixMCP Inspector
Test the MCP server interactively using the official inspector:
pnpx @modelcontextprotocol/inspector node dist/index.jsThis opens a web UI where you can:
- View available tools and their schemas
- Send test notifications with different parameters
- See raw MCP protocol messages
Local Dev Troubleshooting
If you run a local build directly from `dist/index.js` and notifications are not working, make sure the entrypoint is executable:
chmod +x dist/index.jsTesting Notifications
Test the notification system directly without running the MCP server:
# Simple notification (default)
pnpm run test:notification
# With action buttons
pnpm run test:notification actions
# With reply input
pnpm run test:notification reply
# Run all tests
pnpm run test:notification allLicense
Code: MIT License
MCPal Icon & Branding: © 2025 All Rights Reserved. The MCPal logo and icon designs may not be used without permission.
Frequently asked questions
What is MCPal?
MCPal is Friendly Companion MCP. Get notified when your task is finished!
How do I install MCPal?
Open the GitHub repository and follow its README. Most MCP servers are added to your client's MCP config, then called by your agent.
Is MCPal open source?
Yes — it is hosted on GitHub at https://github.com/mjkid221/MCPal and has 1 stars.
Related MCP tools
Browser MCP is a Model Context Provider (MCP) server that allows AI applications to control your browser
Universal AI context generator. Saves thousands of tokens per conversation in Claude Code, Cursor, Copilot, Codex, and more.
📦 Repomix is a powerful tool that packs your entire repository into a single, AI-friendly file. Perfect for when you need to feed your codebase to Large Lan...
MCP server to provide Figma layout information to AI coding agents like Cursor
Put an end to code hallucinations! GitMCP is a free, open-source, remote MCP server for any GitHub project TypeScript-based implementation.
The go-to web for your AI coding agent — local-first search, fetch, crawl & research over MCP. No API keys, no cloud, $0/query. Public beta.
Run your own MCP server? See who uses it and what to fix.
Measure it with TrackMCP