jupyterlab_notifications_extension

JupyterLab extension for sending notifications using the native JupyterLab notification system. External systems and extensions send alerts and status updates that appear in JupyterLab's notification center.

This extension serves as the notification backbone for Stellars JupyterHub Platform for Data Science, allowing administrators to send notification messages to a running JupyterLab server. Each send addresses one server.

Notification types with distinct visual styling provide clear status communication:

Access via command palette for quick manual notification sending:

Interactive dialog with message input, type selection, auto-close timing, and an optional dismiss button:

Key Features:

• REST API for external systems to POST notifications with authentication
• Command palette integration with interactive dialog
• Programmatic command API for extensions and automation
• Six notification types (default, info, success, warning, error, in-progress)
• Configurable auto-close with millisecond precision or manual dismiss
• Action buttons with optional JupyterLab command execution
• Dynamic time-ago indicator showing when each notification was generated
• Best-effort delivery via 30-second polling (the first tab to poll takes it)
• Immediate WebSocket push for instant display (--now / "immediate": true)
• In-memory queue cleared after delivery

Installation

pip install jupyterlab_notifications_extension

Requirements: JupyterLab >= 4.6.0, Python >= 3.10

API Reference

POST /jupyterlab-notifications-extension/ingest

Send notifications to JupyterLab. Requires authentication via an Authorization: token <TOKEN> header. Do not put the token in the URL, where it lands in server and proxy access logs.

Endpoint: POST /jupyterlab-notifications-extension/ingest

Request Body (application/json):

{
  "message": "Your notification message",
  "type": "info",
  "autoClose": 5000,
  "immediate": true,
  "actions": [
    {
      "label": "Click here",
      "caption": "Additional info",
      "displayType": "accent"
    }
  ]
}

Request Parameters:

Action Button Schema:

Note: Clicking any button dismisses the notification. If commandId is provided, the specified JupyterLab command executes before dismissal.

Response (200 OK):

{
  "success": true,
  "notification_id": "notif_1762549476180_1"
}

Error Responses:

• 400 Bad Request - invalid JSON, a body that is not a JSON object, a message that is missing, empty or not a string, an actions that is not a list, an action element that is not an object with a string label, or a number anywhere in the payload that is not finite (NaN, Infinity, or a literal that overflows to one)
• 403 Forbidden - Missing or invalid authentication token
• 500 Internal Server Error - Server-side processing error

Usage Examples

From JupyterLab Extensions

Send notifications programmatically from other extensions:

// Basic notification
await app.commands.execute('jupyterlab-notifications:send', {
  message: 'Operation complete'
});

// Custom type and auto-close
await app.commands.execute('jupyterlab-notifications:send', {
  message: 'Build finished successfully',
  type: 'success',
  autoClose: 3000
});

// With dismiss button
await app.commands.execute('jupyterlab-notifications:send', {
  message: 'Error processing data',
  type: 'error',
  autoClose: false,
  actions: [{ label: 'Dismiss', displayType: 'default' }]
});

// Action button that executes a JupyterLab command
await app.commands.execute('jupyterlab-notifications:send', {
  message: 'Help Available!',
  type: 'info',
  autoClose: false,
  actions: [
    {
      label: 'Open Help',
      commandId: 'iframe:open',
      args: { path: 'local:///welcome.html' },
      displayType: 'accent'
    }
  ]
});

CLI Tool

The jupyterlab-notify command is installed with the extension:

# Basic notification (auto-detects URL from running servers)
jupyterlab-notify -m "Deployment complete" -t success

# With explicit URL (e.g., JupyterHub)
jupyterlab-notify --url "http://127.0.0.1:8888/jupyterhub/user/alice" -m "Hello"

# Persistent warning (no auto-close)
jupyterlab-notify -m "System maintenance in 1 hour" -t warning --no-auto-close

# With dismiss button
jupyterlab-notify -m "Task complete" --action "Dismiss"

# Action button that executes a JupyterLab command
jupyterlab-notify -m "Help Available!" --action "Open Help" \
  --cmd "iframe:open" --command-args '{"path": "local:///welcome.html"}'

# Silent mode (notification center only, no toast)
jupyterlab-notify -m "Background task finished" --auto-close 0

# Immediate display (push now, don't wait for the next poll)
jupyterlab-notify -m "Deploy finished" --now

URL auto-detection: Queries jupyter server list --json to find running servers and constructs the URL from the record's own scheme, host and port, substituting a loopback address for a wildcard bind. When several are running and none matches JUPYTERHUB_SERVICE_PREFIX, it lists them and exits 2 rather than guess. Falls back to JUPYTERHUB_SERVICE_PREFIX, then to http://127.0.0.1:$JUPYTER_PORT (port 8888 when unset). An explicit --url that matches a listed record is re-addressed to that record's own host, so the host printed can differ from the host given: --url http://localhost:8888 for a server bound to 0.0.0.0 is sent to http://127.0.0.1:8888. Any query string or fragment in --url is dropped, because the endpoint path is appended to the URL. Do not put a password in any --url: the user:password@ part is dropped from a --url that names a host, because this tool authenticates by token only, but two malformed shapes keep it - one with no // at all, where the whole user:password@host is read as a path, and one whose password holds an unencoded /, where the host part ends at that slash.

Authentication: JUPYTERLAB_NOTIFY_TOKEN is this tool's own variable and is used for any target, including a remote one. Prefer it over --token, which puts the secret in argv where every local account can read /proc/<pid>/cmdline. The addressed server's own token is sent to the address its runtime record is reached at: 127.0.0.1 for a server bound to 0.0.0.0, and a host that might not be loopback if that server was started with --ip <an address>. The ambient JUPYTERHUB_API_TOKEN / JPY_API_TOKEN / JUPYTER_TOKEN belong to this host, not to the target, so they are sent only to a loopback target.

--now: Pushes the notification instantly to every open JupyterLab tab via WebSocket instead of waiting up to 30 seconds for the next poll (see Immediate Delivery).

cURL

# Localhost - a token is required, as on any other host
curl -X POST http://localhost:8888/jupyterlab-notifications-extension/ingest \
  -H "Content-Type: application/json" \
  -H "Authorization: token YOUR_JUPYTER_TOKEN" \
  -d '{"message": "Build completed", "type": "success"}'

# Localhost - warning that stays until dismissed
curl -X POST http://localhost:8888/jupyterlab-notifications-extension/ingest \
  -H "Content-Type: application/json" \
  -H "Authorization: token YOUR_JUPYTER_TOKEN" \
  -d '{"message": "System maintenance in 1 hour", "type": "warning", "autoClose": false}'

# Remote - requires authentication token
curl -X POST http://jupyterhub.example.com/user/alice/jupyterlab-notifications-extension/ingest \
  -H "Content-Type: application/json" \
  -H "Authorization: token YOUR_JUPYTER_TOKEN" \
  -d '{"message": "Deployment complete", "type": "info"}'

# Immediate display - push now instead of waiting for the next poll
curl -X POST http://localhost:8888/jupyterlab-notifications-extension/ingest \
  -H "Content-Type: application/json" \
  -H "Authorization: token YOUR_JUPYTER_TOKEN" \
  -d '{"message": "Deploy finished", "type": "success", "immediate": true}'

Immediate Delivery

By default a notification waits up to 30 seconds for the frontend's next poll before it appears. Setting immediate (REST/cURL) or passing --now (CLI) pushes it instantly to every open JupyterLab tab over a WebSocket, so it displays the moment it is sent.

• Transport: the frontend keeps a WebSocket open to /jupyterlab-notifications-extension/stream; the server pushes flagged notifications to all connected clients
• Push reaches every connected tab: unlike the poll, the immediate push is delivered to all currently-open tabs at once
• Poll is best-effort: the poll queue is a single-consumer destructive drain - the first tab to poll empties it for all tabs - so a tab whose socket is down at push time is not guaranteed to receive that notification via the poll; the push is an accelerator over a best-effort baseline, not a durable per-client queue
• Keepalive: the socket uses ping/pong to survive proxy idle timeouts (works behind JupyterHub)
• De-duplication: the frontend tracks notification IDs (bounded), so a notification arriving via both the push and the poll is displayed only once
• Reconnect: the socket reconnects with capped exponential backoff, 5 seconds doubling to a 60-second ceiling, and keeps retrying for as long as the tab is open. There is no give-up: while the socket is down this tab competes for the destructive poll queue with every other tab and can miss a --now notification outright, so it warns once in the browser console and keeps trying

Time-Ago Indicator

Each notification displays a relative timestamp (e.g., just now, 5m ago, 2h ago, 3d ago) that updates every 10 seconds while the notification remains visible. Anything under a minute shows just now. The indicator appears below the message when no action buttons are present, or inline with the button bar when buttons exist. The notification center panel also shows time-ago for all listed notifications.

Architecture

No per-recipient addressing: a notification is posted to one server, and the poll queue is taken by the first tab that polls.

Flow: External system POSTs to /jupyterlab-notifications-extension/ingest -> Server queues in memory -> Frontend polls /jupyterlab-notifications-extension/notifications every 30 seconds -> Displays via JupyterLab notification manager -> Clears queue after fetch. Notifications flagged immediate are additionally pushed over a WebSocket (/jupyterlab-notifications-extension/stream) for instant display, deduplicated against the poll by notification ID.

Troubleshooting

Frontend installed but not working:

jupyter server extension list  # Verify server extension enabled

Server extension enabled but frontend missing:

jupyter labextension list  # Verify frontend extension installed

Notifications not appearing: Check browser console for polling errors or verify JupyterLab was restarted after installation.

Uninstall

pip uninstall jupyterlab_notifications_extension

Agent Skill

.agents/skills/jupyterlab-notifications-extension/SKILL.md tells an AI assistant how to drive the jupyterlab-notify CLI. It carries only the rules --help cannot state; the command reference stays in jupyterlab-notify --help.

The skill ships in the repository and in the wheel, which installs it at <sys.prefix>/share/jupyter/agents/skills/jupyterlab-notifications-extension/SKILL.md. No agent reads that directory, and a wheel cannot write into the home directory, so one of the two links below is what makes it readable.

After pip install, with the Python that runs the lab:

mkdir -p ~/.agents/skills && ln -sfn "$(python -c 'import sys; print(sys.prefix)')/share/jupyter/agents/skills/jupyterlab-notifications-extension" ~/.agents/skills/jupyterlab-notifications-extension

From a clone, into Claude Code:

ln -sfn "$PWD/.agents/skills/jupyterlab-notifications-extension" ~/.claude/skills/jupyterlab-notifications-extension

Development

Setup

Requires NodeJS to build the extension. Uses jlpm (JupyterLab's pinned yarn) for package management.

# Install in development mode
python -m venv .venv
source .venv/bin/activate
pip install --editable ".[dev,test]"

# Link extension with JupyterLab
jupyter labextension develop . --overwrite
jupyter server extension enable jupyterlab_notifications_extension

# Build TypeScript
jlpm build

Development workflow

Run jlpm watch in one terminal to auto-rebuild on changes, and jupyter lab in another. Refresh browser after rebuilds to load changes.

jlpm watch           # Auto-rebuild on file changes
jupyter lab          # Run JupyterLab

Cleanup

jupyter server extension disable jupyterlab_notifications_extension
pip uninstall jupyterlab_notifications_extension
# Remove symlink: find via `jupyter labextension list`

Testing

Python tests (Pytest):

pip install -e ".[test]"
pytest -vv -r ap --cov jupyterlab_notifications_extension

Frontend tests (Jest):

jlpm test

Integration tests (Playwright/Galata): See ui-tests/README.md

Packaging

See RELEASE.md for release procedures.