Dismiss once. Gone everywhere.
clrd is a self-hosted notification server. Send an alert with one HTTP call. It lands on your phone, your tablet and the web inbox. Swipe it away on any of them and it clears on all the others.
In early development. The API is settling; expect changes before 1.0.
pixel-9
Nightly backup failed
restic exited 1 after 4m 12s.
web inbox
Nightly backup failed
restic exited 1 after 4m 12s.
Cleared on pixel-9
tablet
Nightly backup failed
restic exited 1 after 4m 12s.
Cleared on pixel-9
How it works
- Send. A script, a monitor or an AI agent posts an alert to a topic.
- It arrives everywhere. Every phone, tablet and browser signed in to that topic shows it.
- Clear it anywhere. Swipe it on the phone, click Clear on the web, or let the sender resolve it when the fault is fixed.
- Every device follows. The server owns the state, so the alert leaves every inbox and every phone tray.
No clocks, no conflicts
Every change is an event with a number. Each device remembers the last number it saw and asks for everything after it when it reconnects.
48211 created backups Nightly backup failed
48212 dismissed backups by pixel-9
A tablet that was off overnight asks for everything after 48211 and drops the alert before you look at it.
More than read and unread
Four states, all synced, all with the device that set them.
- Cleared
- You swiped it away. Gone from every device.
- Acknowledged
- Someone is on it. It stays in the inbox and stops buzzing.
- Resolved
- The sender cleared it because the fault is fixed. Monitors close their own alerts by id.
- Snoozed
- Hidden until a time you pick, then it comes back and notifies again.
Plug it into what you already run
One HTTP call to send. One more to clear it when the problem goes away.
curl
curl -H "Authorization: Bearer $CLRD_TOKEN" \
-d '{"topic":"infra","title":"Disk full",
"message":"web-1 at 95%","priority":4,
"external_id":"disk-web-1"}' \
https://clrd.itsjustin.me/api/v1/notifications
When the disk has room again, the sender clears it everywhere:
curl -X POST -H "Authorization: Bearer $CLRD_TOKEN" \
https://clrd.itsjustin.me/infra/disk-web-1/resolve
AI agents (MCP)
clrd serves MCP at /mcp. Agents can send, list, acknowledge and resolve, and wait_for_reply lets an agent ask you a yes or no question and wait for your tap.
SDKs
Thin clients for Python, JavaScript and Go wrap the same API. The API reference lists every endpoint.
Works with
- Apprise
- Use an
ntfys://URL pointed at your clrd server. - Gatus
- Use the ntfy alert provider. When the endpoint recovers, the alert clears itself.
- Uptime Kuma
- Pick the ntfy notification type and enter your clrd server URL.
- Grafana
- Add a webhook contact point at
/{topic}?template=grafana. Firing alerts open, resolved alerts clear. - Home Assistant
- Use the ntfy integration or a
rest_command. - GitHub Actions
- One curl step at the end of a job.
- Alertmanager
- Point a webhook receiver at
/{topic}?template=alertmanager. Resolved alerts clear on every device. - Any ntfy client
- The ntfy app, CLI and libraries publish and subscribe unchanged.
Already on ntfy?
Point your scripts at clrd and keep going. Same publish endpoint, same headers, same JSON body, same subscribe streams.
curl -H "Authorization: Bearer $CLRD_TOKEN" \
-H "Title: Disk full" -H "Priority: high" \
-d "web-1 at 95%" https://clrd.itsjustin.me/infra
Supported: PUT and POST /{topic}, the Title, Priority, Tags, Click and Actions headers, JSON publish to /, and /json, /sse, /ws and /raw subscriptions with since=.
What changes
| Situation | ntfy | clrd |
|---|---|---|
| You swipe an alert away on one device | The others keep it | Gone everywhere |
| The sender clears an alert | Gone everywhere | Gone everywhere |
| Someone takes an alert | No shared state | Acknowledged, with who |
| A phone was offline | Catches up on messages | Catches up on messages and clears |
Self-host it
One Go binary and one SQLite file. No Redis, no message broker, no account on anyone else's service. clrd makes no outbound calls and sends no telemetry.
A Docker image and compose file ship with the first release. Until then, build from source with Go 1.23 or newer.
git clone https://github.com/thejdubb02/clrd
cd clrd && go build -o clrd ./server/cmd/clrd
CLRD_DB=./clrd.db \
CLRD_BASE_URL=https://notify.example.com \
./clrd
On first start, clrd prints a one-time admin setup link to its log.
The Android app
Built for F-Droid. No Google Play Services, so it runs on GrapheneOS, LineageOS and any phone without Google. It keeps one connection to your server and removes an alert from the tray the moment it is cleared anywhere else.
Not released yet. Follow the GitHub repository for the first build.
Open source
Apache 2.0. The server, the web inbox and the Android app live in one repository, built against one written contract.