Ask your assistant which backups need attention.
BackupSentinel runs an MCP server. Connect Claude Code, Claude Desktop, Cursor or another MCP client with an API key, and it can read the same queue, clients, backups and reports you see in the dashboard.
# Settings → API keys → Create key claude mcp add --transport http backupsentinel \ https://app.backupsentinel.io/api/mcp \ --header "Authorization: Bearer bsk_live_…" # then ask, in Claude Code: Which backups need attention this morning?
One key, one command.
- An owner or admin creates a key under Settings → API keys and chooses its access: Read, or Read and act.
- The key is shown once. Copy the
claude mcp addcommand shown with it, or add the URL and header to your client’s settings. - Ask in your own words. The assistant picks the tools; every answer comes from your workspace at that moment.
There is also a morning_check prompt: what needs a technician, grouped by client, worst first, and what is already acknowledged.
For clients set up with JSON
{
"mcpServers": {
"backupsentinel": {
"type": "http",
"url": "https://app.backupsentinel.io/api/mcp",
"headers": { "Authorization": "Bearer bsk_live_…" }
}
}
}The key is sent as a header, so use a client that can send one, such as Claude Code, Claude Desktop or Cursor.
What it can read, with any key.
The same statuses as the dashboard (Healthy, Warning, Failed, Missing, Waiting, Paused), the same order, and a link to the matching page with each answer.
| Tool | What it does |
|---|---|
workspace_summary | Backups per status, open problems, Inbox emails waiting for review, clients by status. |
needs_attention | The Needs attention queue, worst first, with how long each problem has lasted and earlier fixes on the same backup. |
list_clients, get_client | Clients with their SLA tier and backups per status; one client's backups, contacts and devices. |
list_backups, get_backup | Backups by status, client or product; one backup's interval, grace, last 20 reports and open alerts. |
list_inbox, get_inbox_email | Report emails to review, what the rules read from each, and the start of the body as text. |
client_report | A client's month or quarter: success rate, missing reports, incidents and time to recover. |
list_channels | Alert channels, which clients they cover and what each Email channel sends. |
What it can change, with Read and act.
A key created with Read and act also gets these tools; a Read key does not see them, and a call by name is refused. Each change follows the dashboard’s own rules and is written to the activity log as API key "name" (MCP), so you can tell what the assistant did from what a person did.
| Tool | What it does |
|---|---|
acknowledge_alert | Marks an alert as being handled, as Acknowledge does in Needs attention. |
snooze_alert | Hides an alert for 15 minutes to 7 days. It comes back sooner if the backup fails again. |
resolve_alert | Closes an alert with a note. Channels that received it get a recovery notice. |
report_run | Records a backup result, exactly like the Report API. |
pause_backup, resume_backup | Pauses monitoring of one backup, until a date or until resumed. |
What it does not do.
- A key only ever sees and changes its own workspace.
- It never returns device logins, webhook URLs, API keys, billing or team members. Channel addresses are masked, and report emails stop after 2,000 characters.
- It cannot create or delete clients, backups, channels or keys.
- It signs in with an API key only. The claude.ai web app needs an OAuth sign-in, which is not available yet.
- 60 requests a minute per key, shared with the Report API. A suspended workspace gets an error.
Ask about your own backups.
Start the trial, add a client, create a key and connect your client.