BackupDrill plugin

Back up a Supabase project to a bucket you own and prove the backup restores, by asking your coding agent. The plugin is one skill that teaches the agent to use backupdrill, the open-source (MIT) CLI behind backupdrill.com:

• estimate measures the database and Storage files and projects the Supabase egress cost of backing them up.
• backup dumps the Postgres database with pg_dump, plus Storage files once you add their S3 keys, into your S3, R2 or B2 bucket, with a checksummed manifest.json and a RECOVERY.md in every snapshot.
• drill restores a snapshot into a throwaway Postgres in Docker, checks that the tables and their data came back, and checksums a sample of the Storage files.
• restore puts a snapshot into a new, empty Supabase project, after a dry run and only once you confirm the target.

The agent runs every command as npx -y backupdrill@2.10.0, so the CLI version stays fixed.

What you need

• A shell the agent can use, such as Claude Code. In a chat with no terminal, the agent can explain the commands but not run them.
• Node.js 20 or newer.
• pg_dump and pg_restore version 17, which handle Supabase's Postgres 15 to 17. On macOS: brew install libpq@17 (plain libpq is now 18, whose dumps can contain SQL that Postgres 17 rejects).
• Docker, for drill.
• A bucket on AWS S3, Cloudflare R2 or Backblaze B2, with a key that can list, read and write objects in it.

Install

In Claude Code:

/plugin marketplace add backupdrill/cli
/plugin install backupdrill@backupdrill

Or from a terminal:

claude plugin install backupdrill --marketplace backupdrill/cli

In Pi, the same skill comes with the npm package:

pi install npm:backupdrill

Things to ask

• "Back up my Supabase project to my R2 bucket."
• "How much egress would a daily backup of this project add on the Pro plan?"
• "Run a restore drill on the latest backup and tell me whether it passed."
• "Set up a weekly backup in GitHub Actions."
• "Restore last night's snapshot into my new staging project. Dry run first."

Your credentials

The skill keeps your connection string and bucket keys in .env.backupdrill in your project root and adds that file to .gitignore. It creates the file with placeholders and asks you to fill it in your editor. The agent is told never to open or print the file, never to ask for keys in the chat, and to load the file only for the command that needs it.

What runs and where your data goes

Everything runs on your machine:

• npx downloads the backupdrill package from the npm registry the first time.
• The CLI connects to your Supabase project's database, to its Storage if you add Storage S3 keys, and to your bucket. restore also connects to the target project you name.
• drill starts a temporary Postgres container in Docker. The first time, Docker downloads the Postgres image from Docker Hub.
• Nothing goes to BackupDrill. There is no account, telemetry or license check.

What the agent reads while it works, such as command output with table names, row counts and file names, goes to your model provider like the rest of your conversation. The CLI itself never deletes objects in your bucket.

Troubleshooting

• Missing required config: the message names the missing setting. Fill that line in .env.backupdrill.
• A pg_dump version error: install the version 17 PostgreSQL client and set BACKUPDRILL_PG_DUMP to its pg_dump.
• Can't connect to the database: use the Session pooler string from the Connect button in the Supabase dashboard, not the Direct one. Resume the project if it's paused.
• drill can't start a container: start Docker.

The CLI's README covers every option, and backupdrill.com/docs/cli walks through the same steps.

Support

Open an issue at github.com/backupdrill/cli/issues or email support@backupdrill.com. Privacy policy: backupdrill.com/privacy.

License

MIT