Backup MCPPostgreSQL & MongoDB

Backup MCP: a fresh backup before every migration

Before your AI agent runs a migration, it starts a real Backup job and waits for it to succeed. Your restore point is minutes old, and the agent never reads or changes your data.

PostgreSQL & MongoDBOn-demand backup jobsNo SQL accessWrite access is opt-in
Example conversation. Tool results are shortened.

A restore point, one sentence away

Before you run a migration, ask your agent to back up the production database. It starts the backup and checks the job until it reports success. Only then does it move on to the migration.

  • "Back up the prod database, then run the migration"
  • The agent finds the database by name, starts the job, waits until it finishes
  • A failed backup stops the plan before anything destructive runs
See the workflow in the docs

Backup tools, one job each

Each tool calls the same Backup API the dashboard uses. Your agent only sees the ones you granted.

Find the right database

Lists the databases connected to Backup, so the agent picks the right one by name.

Find the right database

Lists the databases connected to Backup, so the agent picks the right one by name.

Lists the databases connected to Backup in your organization, so the agent picks the right one by name instead of guessing an id. Tool: backup_list_datasources. Needs Backup read.

Start a backup

Starts a backup now, exactly like Run backup now in the dashboard.

Start a backup

Starts a backup now, exactly like Run backup now in the dashboard.

Starts an on-demand backup, exactly like Run backup now in the dashboard, and returns the job id. Optionally takes a specific attached policy; otherwise the highest-priority enabled one is used. Tool: backup_run. Needs Backup write.

Wait until it succeeds

Reports the job status, so the agent waits until the backup succeeds.

Wait until it succeeds

Reports the job status, so the agent waits until the backup succeeds.

Reports where a job is (queued, running, success or failed) and whether it has finished. The agent checks it until the job is done. Tool: backup_get_status. Needs Backup read.

Prove it restores

Restore-tests the backup in a throwaway database. Yours is never touched.

Prove it restores

Restore-tests the backup in a throwaway database. Yours is never touched.

Restores a backup into a throwaway database, validates it, and throws it away: the same restore test as Test now in the dashboard. Never touches your own database. Tools: backup_run_restore_check, backup_get_restore_check. Needs the separate Run restore checks permission.

Know your track record

Answers "when was the last good backup?" from your backup and restore-test history.

Know your track record

Answers "when was the last good backup?" from your backup and restore-test history.

"When was the last good backup?" and "has prod ever been restore-tested?" answered from your backup history and recent restore-test results. Tools: backup_list_jobs, backup_restore_check_overview. Needs Backup read.

Guardrails

Your agent gets your limits, nothing more

Backups, not SQL access

Many database MCP servers hand the model a query console. This one doesn't. Your agent can start and monitor backups, but there's no tool that runs queries, reads rows or downloads a dump.

Write access is a separate permission

Listing and status need Backup read. Starting a backup or a restore check each needs its own permission, selected on the consent screen and enforced by Backup itself.

Same rules as the dashboard

Every call runs through your Norcube role and your organization's billing status. A run from the agent is the same manual run you'd start yourself, recorded in the audit log.

Revoke in one click

Account settings → Connected apps lists every client you've connected. Revoking one cuts it off on its next tool call.

Common questions

Can the agent read or change my data?
No. The Backup MCP tools list your databases, start backups and report job status. There's no tool that runs SQL, reads table contents or downloads a backup.
Does an agent-triggered backup cost extra?
No. It's an ordinary manual backup job and counts against your Backup plan the same way as clicking Run backup now.
Will the backup lock my database?
No. Backups use pg_dump or mongodump from an isolated server. Reads and writes keep working while the dump runs; on PostgreSQL only schema changes wait until it finishes.
What does the database need before the agent can back it up?
Stored credentials and at least one enabled backup policy attached to it, the same as a manual run in the dashboard. If either is missing, backup_run returns an error explaining what to fix.
Can the agent test that my backup restores?
Yes, with the Run restore checks permission. It restores the backup into a throwaway database (Norcube-managed, or a fresh database on your scratch server), validates it and reports whether it passed. Your own database is never touched.
Can the agent restore a backup into my database?
No. A real restore can overwrite data, so restores and downloads stay in the dashboard and the Norcube CLI.
How long does the agent wait?
As long as the job runs. Small databases usually finish within minutes; large ones can take much longer. The agent keeps calling backup_get_status until the job reports success or failed.

Never migrate without a fresh backup

Connect a database to Backup, add the Norcube MCP server to your agent, and make "back up first" part of every deploy.

  • PostgreSQL & MongoDB
  • No SQL access
  • Backup write is opt-in
  • Audited
// tick. tick. tick.
0 ticks since founding
Coming soon

Join the PromptHub waitlist

PromptHub is in active development. Leave your email and we'll invite you as soon as it opens, with a free tier and no credit card.

One email when access opens. No newsletter, no spam.