Your agent.
The same account.

Connect

1. Sign in, open your dashboard and create an API key. It is shown once; store it in your agent's secret manager.
2. Set your MCP client's HTTP server URL to https://YOUR-DEPLOYMENT/api/mcp and send Authorization: Bearer bs_….
3. Initialize, list tools, then call list_sources. Revoke the key on the dashboard when you no longer need it.

The endpoint speaks stateless JSON-RPC over HTTP POST, protocol version 2025-03-26. initialize and ping work without a key; tools/list and tools/call need one. Notifications return 202. There is no GET stream or server-initiated event; GET returns the server's name and protocol. A failed tool call returns its error as text with isError: true.

Tools

list_sources

List your sources, with each one's host link and execution state. none

add_source

Save a database, server, service or computer. Passwords are ignored and saving starts nothing. name, kind, storageMode; engine, host, user, database for a database

list_destinations

List your storage, its probeStatus and usage. none

add_destination

Save storage. New storage is unverified until a host reaches it. name, vendor, bucket, place?

start_backup

Queue a PostgreSQL backup on the source's main host, outside any plan. Returns the job, not a finished backup. sourceId, requestId?

list_backup_plans

List backup plans with schedule, storage, the host each runs on, next run and last backup. none

save_backup_plan

Create a plan, or update one by id: source, storage, schedule and the host that runs it. Saving does not run a backup. sourceId, name, id?, destinationIds?, scheduleMinutes?, firstRunAt?, enabled?, coordinatorId?, selection?, dumpStreams?

remove_backup_plan

Remove a plan. Its past backups and archives are kept. planId

run_backup_plan

Queue a plan's run now on its host, with its storage and settings. Returns the job. planId, requestId?

cancel_backup

Stop a queued or running job. A running host stops within about 15 seconds and keeps nothing from that run. jobId

list_backups

Up to 100 recent backup jobs with status and the host's reported result. sourceId?

run_restore_check

Queue an isolated restore check of one backup on the host that made it. sourceId, backupId, requestId?

start_restore

Always returns unavailable (503). Recovery into a separate database is manual. sourceId, backupId, location

Destination vendors are r2, b2, s3, hetzner, wasabi, gcs, azure, digitalocean, scaleway, s3compat and ssh. For ssh, place is ssh://user@host[:port] and bucket is an absolute directory. Poll list_backups, or GET /api/v1/restore-checks for restore checks, to see results. Queued is not finished.