ntfy-mcp

ntfy-mcp

Enables AI agents to send push notifications to your phone through ntfy, with built-in security controls to prevent data exfiltration. It exposes a single tool notify_user for notifying when tasks complete or need attention.

Category
Visit Server

README

ntfy-mcp

Python MCP License: MIT

ntfy-mcp is a small, security-focused MCP server that lets AI agents send push notifications through ntfy.

It is designed for the practical moments when a long-running agent needs your attention:

  • a build or test run finishes
  • tests fail while you are away
  • an approval is needed
  • a PR is ready for review
  • a task fails and needs human attention

The server exposes one MCP tool:

notify_user

Under the hood it sends an HTTPS POST to an operator-configured ntfy topic. The model can write the notification content, but it cannot choose arbitrary ntfy servers, arbitrary topics, arbitrary headers, attachments, or action buttons.

Why This Exists

Coding agents are increasingly good at running for several minutes without help. The missing piece is often human attention. Desktop notifications tied to one app are easy to miss, and polling a terminal is tedious.

ntfy-mcp gives MCP-capable clients a tiny, boring, auditable path to notify you on your phone without turning ntfy into an unbounded exfiltration channel.

Security Model

This project treats notification delivery as a security boundary, not as a thin wrapper around ntfy.

The MCP model is considered untrusted input. It must not control:

  • the ntfy base URL
  • arbitrary topics
  • auth headers
  • arbitrary request headers
  • file attachments
  • action buttons

MVP controls:

  • NTFY_TOPIC is required.
  • NTFY_ALLOWED_TOPICS defaults to NTFY_TOPIC.
  • Requested topics must be in the allow-list.
  • Topics must match a conservative letters, numbers, _, - format.
  • NTFY_BASE_URL must use HTTPS and cannot include credentials.
  • Bearer auth only comes from NTFY_TOKEN.
  • Priority is constrained to 1 through 5.
  • Long messages are truncated before sending.
  • Obvious secret-like content is rejected before sending.
  • click_url must be http:// or https://.
  • Only a fixed set of ntfy headers is sent.
  • Full notification messages are not logged by default.
  • NTFY_DRY_RUN=true validates and returns success without network I/O.

Secret detection is intentionally basic. It blocks obvious private keys, GitHub tokens, OpenAI-style tokens, JWT-looking tokens, AWS access keys, and common token=, password=, api_key= style assignments. It is a guardrail, not a replacement for careful tool permissions.

Quickstart

Install dependencies:

uv sync

Choose a long random topic:

openssl rand -hex 24

Run a dry-run MCP client smoke test:

NTFY_TOPIC="replace-with-long-random-topic" \
NTFY_ALLOWED_TOPICS="replace-with-long-random-topic" \
NTFY_DRY_RUN=true \
uv run python examples/mcp-client-smoke.py

Run the STDIO MCP server directly:

NTFY_TOPIC="replace-with-long-random-topic" \
NTFY_ALLOWED_TOPICS="replace-with-long-random-topic" \
uv run python -m ntfy_mcp.server

The server waits for MCP messages on STDIO. Use an MCP client to call notify_user.

Subscribe In The ntfy App

  1. Install the ntfy mobile app.
  2. Add a subscription.
  3. Use https://ntfy.sh unless you self-host ntfy.
  4. Enter your long random topic.
  5. If your ntfy server requires auth, configure a token on the server and set NTFY_TOKEN for this MCP server.

Avoid short or guessable topics. Public ntfy topics are effectively bearer secrets unless protected by server-side access control.

Test ntfy Directly

Before wiring an MCP client, verify ntfy delivery:

curl -fsS \
  -H "Title: ntfy-mcp test" \
  -H "Priority: 3" \
  -d "hello from ntfy-mcp" \
  "https://ntfy.sh/$NTFY_TOPIC"

For protected topics:

curl -fsS \
  -H "Authorization: Bearer $NTFY_TOKEN" \
  -H "Title: ntfy-mcp test" \
  -d "hello from ntfy-mcp" \
  "https://ntfy.sh/$NTFY_TOPIC"

Codex Setup

Add a server entry to ~/.codex/config.toml:

[mcp_servers.ntfy]
command = "uv"
args = ["--directory", "/absolute/path/to/ntfy-mcp", "run", "python", "-m", "ntfy_mcp.server"]
enabled = true
enabled_tools = ["notify_user"]

[mcp_servers.ntfy.env]
NTFY_BASE_URL = "https://ntfy.sh"
NTFY_TOPIC = "replace-with-long-random-topic"
NTFY_ALLOWED_TOPICS = "replace-with-long-random-topic"
NTFY_SOURCE = "codex"
NTFY_DEFAULT_PRIORITY = "3"
NTFY_MAX_MESSAGE_LENGTH = "1800"

If your topic requires auth:

NTFY_TOKEN = "replace-with-ntfy-access-token"

Keep tokens in local config or a secret manager. Do not commit them.

Claude Desktop Setup

Add a server entry to Claude Desktop's MCP configuration:

{
  "mcpServers": {
    "ntfy": {
      "command": "uv",
      "args": [
        "--directory",
        "/absolute/path/to/ntfy-mcp",
        "run",
        "python",
        "-m",
        "ntfy_mcp.server"
      ],
      "env": {
        "NTFY_BASE_URL": "https://ntfy.sh",
        "NTFY_TOPIC": "replace-with-long-random-topic",
        "NTFY_ALLOWED_TOPICS": "replace-with-long-random-topic",
        "NTFY_SOURCE": "claude",
        "NTFY_DEFAULT_PRIORITY": "3",
        "NTFY_MAX_MESSAGE_LENGTH": "1800"
      }
    }
  }
}

Add NTFY_TOKEN inside env only if the ntfy topic requires bearer auth.

Tool Input

notify_user accepts:

Field Type Notes
title string Required. Sent as the ntfy title.
message string Required. Truncated to NTFY_MAX_MESSAGE_LENGTH.
severity info, success, warning, error Required by schema default.
priority integer Optional. Must be 1 through 5.
tags list of strings Optional. Conservative tag aliases only.
click_url string Optional. Must start with http:// or https://.
topic string Optional. Must be in NTFY_ALLOWED_TOPICS.

Severity adds a default tag:

Severity Default tag
info information_source
success white_check_mark
warning warning
error rotating_light

Priority behavior:

  • Explicit priority wins if it is 1 through 5.
  • info and success use NTFY_DEFAULT_PRIORITY.
  • warning defaults to at least 4 when no explicit priority is supplied.
  • error defaults to 5 when no explicit priority is supplied.

Example Prompts

  • "Notify me when the build finishes."
  • "Notify me only if tests fail."
  • "Notify me when you need approval."
  • "Notify me when the PR is ready for review."

Environment Variables

Variable Required Default Description
NTFY_BASE_URL No https://ntfy.sh HTTPS ntfy server base URL.
NTFY_TOPIC Yes None Default topic. Must pass topic validation.
NTFY_ALLOWED_TOPICS No NTFY_TOPIC Comma-separated allow-list.
NTFY_TOKEN No None Bearer token used for the Authorization header.
NTFY_DEFAULT_PRIORITY No 3 Default priority for info and success.
NTFY_MAX_MESSAGE_LENGTH No 1800 Maximum message length before truncation.
NTFY_SOURCE No agent Operator-controlled source label.
NTFY_DRY_RUN No false Validate and return success without network I/O.

Development

uv run pytest
uv run ruff check .

Dry-run the full MCP path:

NTFY_TOPIC="ntfymcp_smoke_topic_123" \
NTFY_ALLOWED_TOPICS="ntfymcp_smoke_topic_123" \
NTFY_DRY_RUN=true \
uv run python examples/mcp-client-smoke.py

Limitations

  • STDIO first.
  • Remote Streamable HTTP mode is future work.
  • Attachments are intentionally omitted from the MVP.
  • Action buttons are intentionally omitted from the MVP.
  • Secret detection is basic and intentionally conservative.
  • No local rate limiting yet.

Roadmap

  • Optional local rate limiting and duplicate suppression.
  • Streamable HTTP mode with authentication guidance.
  • Stronger configurable secret scanning.
  • Metadata-only audit logging.
  • Notification templates for common agent states.
  • Packaged releases and signed artifacts.

License

MIT. See LICENSE.

Recommended Servers

playwright-mcp

playwright-mcp

A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.

Official
Featured
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.

Official
Featured
Local
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.

Official
Featured
Local
TypeScript
VeyraX MCP

VeyraX MCP

Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.

Official
Featured
Local
graphlit-mcp-server

graphlit-mcp-server

The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.

Official
Featured
TypeScript
Kagi MCP Server

Kagi MCP Server

An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.

Official
Featured
Python
E2B

E2B

Using MCP to run code via e2b.

Official
Featured
Neon Database

Neon Database

MCP server for interacting with Neon Management API and databases

Official
Featured
Exa Search

Exa Search

A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.

Official
Featured
Qdrant Server

Qdrant Server

This repository is an example of how to create a MCP server for Qdrant, a vector search engine.

Official
Featured