ntfy-mcp-server
An MCP (Model Context Protocol) server designed to interact with the ntfy push notification service. It enables LLMs and AI agents to send notifications to your devices with extensive customization options.
Documentation
Tools
Four tools covering the ntfy publish/subscribe surface — message lifecycle (publish, manage, fetch) plus an emoji-tag lookup that feeds the publish tool's `tags` field:
| Tool Name | Description |
|---|---|
| `ntfy_publish_message` | Send or update a push notification on an ntfy topic. |
| `ntfy_manage_message` | Clear or delete a previously-sent notification by `sequence_id`. |
| `ntfy_fetch_messages` | Poll cached messages from one or more topics with optional filters. |
| `ntfy_search_emoji_tags` | Look up ntfy emoji tag short codes for use in `tags`. |
`ntfy_publish_message`
Send or update a push notification on an ntfy topic. Topics are created on first publish — treat the topic name as a secret because anyone who knows it can publish or subscribe.
- Full publish-parameter coverage — `title`, `priority` (1–5), `tags`, `click`, `attach`, `icon`, `filename`, `markdown`, `delay`, `email`, `call`, `cache`, `firebase`
- Up to three discriminated action buttons (`view`, `broadcast`, `http`, `copy`) per message
- Update or replace previously-sent messages by passing the original `sequence_id`
- Per-call `base_url` override that forwards credentials only when the override matches a registered server (`NTFY_BASE_URL` or an `NTFY_SERVERS` entry); otherwise the request goes out unauthenticated, so credentials never leak to alternate hosts
- Publishes carrying `email`, `call`, or a `broadcast` / `http` action button ask the user to confirm the specific recipient or target first — the call comes back requesting that confirmation, and the message goes out only when it is reissued with the answer attached
`ntfy_manage_message`
Clear (mark read & dismiss) or delete a previously-sent ntfy notification by `sequence_id`. Append-only — the original message stays in cache, and a `message_clear` / `message_delete` event is emitted to subscribers. Idempotent.
- Every call asks the user to confirm the topic, `sequence_id`, and operation before the event fires — the first call comes back requesting that confirmation, and declining fails the call with `consent_declined`
`ntfy_fetch_messages`
Poll cached messages from one or more topics with optional filters. Returns a snapshot, not a live stream — use it to confirm delivery, replay missed alerts, or audit topic activity.
- Comma-separated multi-topic queries (e.g. `alerts,backups,phil_alerts`)
- Filter by `since` (duration / timestamp / message ID / `all` / `latest`), `priority`, `tags`, `id`, `title`, `message`, scheduled-only
- Default window `10m`, default limit 20 messages per response, hard cap 100 — over-limit windows keep the newest `limit` messages, listed oldest-first
- Long bodies truncated to ~500 chars with `messageTruncated` reporting the dropped count; refetch with a message `id` to read that one in full
`ntfy_search_emoji_tags`
Substring search over the bundled ntfy emoji-tag reference. Returns the `tag` strings ready to plug into `ntfy_publish_message`'s `tags` field. Without a query, returns the first slice of the full reference; `offset` pages through matches beyond the `limit` cap.
Resources and prompts
| Type | Name | Description |
|---|---|---|
| Resource | `ntfy://{topic}` | Snapshot of a topic — latest 20 messages from the past 1 hour, plus the topic's browser URL. Same normalized message shape as `ntfy_fetch_messages`: ISO 8601 timestamps and ~500-char body truncation. |
`ntfy_fetch_messages` covers the same topic data with custom windows and filters when the resource's fixed defaults aren't enough.
Features
Built on `@cyanheads/mcp-ts-core`:
- Declarative tool and resource definitions — single file per primitive, framework handles registration and validation
- Typed error contracts via `ctx.fail(reason, …)` plus framework error factories (`forbidden`, `notFound`, `validationError`, …)
- Pluggable auth: `none`, `jwt`, `oauth`
- Swappable storage backends: `in-memory`, `filesystem`, `Supabase`, `Cloudflare KV/R2/D1`
- Structured logging with optional OpenTelemetry tracing
- STDIO and Streamable HTTP transports
ntfy-specific:
- Wraps ntfy's HTTP API with retry-aware client (`withRetry` + per-request timeout)
- Per-server scoped auth — credentials are bound to each registered base URL (`NTFY_BASE_URL` or per-entry under `NTFY_SERVERS`); per-call `base_url` overrides forward auth only when the override matches a registered server, and go out unauthenticated otherwise
- User confirmation before side effects that leave the notification drawer — a clear/delete, or a publish carrying `email`, `call`, or a `broadcast` / `http` action button. The tool returns a confirmation request naming the exact target, and acts only on the reissued call that carries an approval. Enforced on both STDIO and Streamable HTTP; a client that cannot present the prompt gets an error instead of an unasked side effect
- Optional SSRF guard on `base_url` overrides (`NTFY_BLOCK_PRIVATE_HOSTS`) — resolves the host and blocks every reserved destination it answers on (loopback, RFC 1918, RFC 6598 mesh space, link-local, and the IPv6 equivalents), then refuses redirects, with registered servers exempt
- Bundled emoji-tag reference, regenerated from upstream `docs/ntfy/emojis.md` via `scripts/build-emoji-tags.ts`
- Mutually-exclusive auth modes (bearer token *or* basic auth) validated at config-load time
Getting started
Add the following to your MCP client configuration file. Public ntfy.sh works out of the box without an account; for protected topics, generate an access token at .
{
"mcpServers": {
"ntfy-mcp-server": {
"type": "stdio",
"command": "bunx",
"args": ["ntfy-mcp-server@latest"],
"env": {
"MCP_TRANSPORT_TYPE": "stdio",
"MCP_LOG_LEVEL": "info",
"NTFY_DEFAULT_TOPIC": "your-topic-name"
}
}
}
}Or with Docker:
{
"mcpServers": {
"ntfy-mcp-server": {
"type": "stdio",
"command": "docker",
"args": [
"run", "-i", "--rm",
"-e", "MCP_TRANSPORT_TYPE=stdio",
"-e", "NTFY_DEFAULT_TOPIC=your-topic-name",
"ghcr.io/cyanheads/ntfy-mcp-server:latest"
]
}
}
}For Streamable HTTP, set the transport and start the server:
MCP_TRANSPORT_TYPE=http MCP_HTTP_PORT=3010 NTFY_DEFAULT_TOPIC=your-topic bun run start:http
# Server listens at http://127.0.0.1:3010/mcpPrerequisites
- Bun v1.3.11 or higher (or Node.js v24+).
- A topic name on an ntfy server. Public `ntfy.sh` requires no account; self-hosted instances and protected topics may need a bearer token or basic-auth credentials.
Installation
1. Clone the repository:
git clone https://github.com/cyanheads/ntfy-mcp-server.git2. Navigate into the directory:
cd ntfy-mcp-server3. Install dependencies:
bun install4. Configure environment:
cp .env.example .env
# edit .env and set NTFY_DEFAULT_TOPIC (and auth, if needed)Configuration
| Variable | Description | Default | |
|---|---|---|---|
| `NTFY_SERVERS` | JSON array of `{ baseUrl, authToken? \ | authUsername?+authPassword? }` entries — one per ntfy server. First entry is the default base. Auth is scoped to the entry's `baseUrl`; per-call `base_url` overrides that match a registered base forward that server's auth. Use this when you need more than one authenticated server in a single process; it takes precedence over the single-server vars below. | — |
| `NTFY_BASE_URL` | Single-server shorthand — base URL of the ntfy server (no trailing slash). Used when `NTFY_SERVERS` is unset. | `https://ntfy.sh` | |
| `NTFY_DEFAULT_TOPIC` | Topic used when a tool call omits `topic`. | — | |
| `NTFY_AUTH_TOKEN` | Bearer access token (`tk_…`) for the single-server shorthand. Mutually exclusive with `NTFY_AUTH_USERNAME` / `NTFY_AUTH_PASSWORD`. | — | |
| `NTFY_AUTH_USERNAME` | Basic-auth username for the single-server shorthand — required together with `NTFY_AUTH_PASSWORD`. | — | |
| `NTFY_AUTH_PASSWORD` | Basic-auth password for the single-server shorthand — required together with `NTFY_AUTH_USERNAME`. | — | |
| `NTFY_REQUEST_TIMEOUT_MS` | Per-request HTTP timeout in milliseconds. | `15000` | |
| `NTFY_MAX_RETRIES` | Max retry attempts for transient upstream failures (5xx, network, 429). | `3` | |
| `NTFY_BLOCK_PRIVATE_HOSTS` | When `true`, a per-call `base_url` override must resolve to a public address, and its redirects are not followed. Servers registered under `NTFY_SERVERS` / `NTFY_BASE_URL` are exempt, so a deliberate LAN target still works. Turn it on where callers you don't control can reach the server. | `false` | |
| `MCP_TRANSPORT_TYPE` | Transport: `stdio` or `http`. | `stdio` | |
| `MCP_SESSION_MODE` | HTTP session model: `auto`, `stateful`, or `stateless`. `auto` resolves to `stateful`. Keep it stateful — the consent prompt on destructive and outbound calls is a multi-round-trip request that a 2025-era HTTP client can only complete over a live session. | `auto` (→ `stateful`) | |
| `MCP_HTTP_HOST` | HTTP host. | `127.0.0.1` | |
| `MCP_HTTP_PORT` | HTTP port. | `3010` | |
| `MCP_HTTP_ENDPOINT_PATH` | HTTP endpoint path. | `/mcp` | |
| `MCP_AUTH_MODE` | Auth mode: `none`, `jwt`, or `oauth`. | `none` | |
| `MCP_LOG_LEVEL` | Log level (RFC 5424). | `info` | |
| `LOGS_DIR` | Directory for file-based logs (Node only; ignored on Workers). | `./logs` | |
| `OTEL_ENABLED` | Enable OpenTelemetry instrumentation (spans, metrics, completion logs). | `false` |
See `.env.example` for the full list of optional overrides.
Running the server
Local development
- Build and run:
# One-time build
bun run rebuild
# Run the built server
bun run start:stdio
# or
bun run start:http- Run checks and tests:
bun run devcheck # Lint, format, typecheck, security, changelog sync
bun run test # Vitest test suite
bun run lint:mcp # Validate MCP definitions against specDocker
docker build -t ntfy-mcp-server .
docker run --rm -e NTFY_DEFAULT_TOPIC=your-topic -p 3010:3010 ntfy-mcp-serverThe Dockerfile defaults to HTTP transport, stateful session mode, and logs to `/var/log/ntfy-mcp-server`. OpenTelemetry peer dependencies are installed by default — build with `--build-arg OTEL_ENABLED=false` to omit them.
Project structure
| Directory | Purpose |
|---|---|
| `src/index.ts` | `createApp()` entry point — registers tools and resources, initializes services. |
| `src/config` | Server-specific environment variable parsing (`NTFY_*`) with Zod. |
| `src/mcp-server/tools` | Tool definitions (`*.tool.ts`). |
| `src/mcp-server/resources` | Resource definitions (`*.resource.ts`). |
| `src/services/ntfy` | ntfy HTTP client, types, and error classifier. |
| `src/services/emoji-tags` | Bundled emoji short-code reference and lookup service. |
| `docs/ntfy` | Mirrored upstream ntfy API docs (pinned commit in `SOURCES.md`). |
| `tests/` | Unit and integration tests mirroring `src/`. |
Development guide
See `CLAUDE.md` for development guidelines and architectural rules. The short version:
- Handlers throw, framework catches — no `try/catch` in tool logic
- Use `ctx.log` for request-scoped logging, `ctx.state` for tenant-scoped storage
- Wrap external API calls: validate raw → normalize to domain type → return output schema; never fabricate missing fields
- Per-tool `errors[]` contracts stay inline — repetition is intended for locality
Contributing
Issues and pull requests are welcome. Run checks and tests before submitting:
bun run devcheck
bun run testLicense
Apache-2.0 — see LICENSE for details.
Frequently asked questions
What is ntfy-mcp-server?
ntfy-mcp-server is An MCP (Model Context Protocol) server designed to interact with the ntfy push notification service. It enables LLMs and AI agents to send notifications to your devices with extensive customization options.
How do I install ntfy-mcp-server?
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 ntfy-mcp-server open source?
Yes — it is hosted on GitHub at https://github.com/cyanheads/ntfy-mcp-server and has 14 stars.
Related MCP tools
🤖 A visualization mcp contains 25+ visual charts using @antvis. Using for chart generation and data analysis. TypeScript-based implementation.
A Unified MCP Server Management App TypeScript-based implementation. Trusted by 800+ developers. Trusted by 800+ developers.
Enable any LLM (e.g. Claude) to interactively debug any language for you via MCP and a VS Code Extension TypeScript-based implementation.
Add Obsidian integrations like semantic search and custom Templater prompts to Claude or any MCP client. TypeScript-based implementation.
An MCP client for Neovim that seamlessly integrates MCP servers into your editing workflow with an intuitive interface for managing, testing, and using MCP s...
Plugin for JADX to integrate MCP server Java-based implementation. Trusted by 600+ developers. Trusted by 600+ developers.
Run your own MCP server? See who uses it and what to fix.
Measure it with TrackMCP