Bubble CLI — mirror your app's data into a local SQLite database

Showcase
1

image
image504×193 1.81 KB

Hey everyone,

I built a small command-line tool that mirrors a Bubble app’s data into a local SQLite database via the Data API. Sharing it here in case it’s useful — and to get feedback from anyone who tries it.

What it does

You point it at a Bubble app, run bubble scan, and it pulls your types and fields from the /meta endpoint. Then bubble pull downloads the records into a SQLite file sitting next to your project. From there you can query it with anything that speaks SQL — DB Browser, DBeaver, a Python notebook, Metabase, whatever.

A few things it handles

  • Schema discovery with a fallback: if /meta doesn’t return types (older apps, certain configs), it samples a record per type and infers the fields.

  • Two sync modes — full re-fetch or incremental (only records with Modified Date newer than the last sync).

  • Schema diff on every scan, so you see which types/fields were added or removed since last run.

  • Per-folder projects — config lives in bubble.json, so you can keep multiple apps side by side without them stepping on each other.

  • Interactive menu when you run bubble with no args, plus scriptable subcommands for cron/CI.

  • Bilingual UI (English / Português), auto-detected from your locale.

Use cases I had in mind

  • Local backups of production data without paying for a separate backup service.

  • Ad-hoc analytics — JOINs, GROUP BYs and window functions that aren’t really practical inside Bubble’s search.

  • Feeding data into BI tools, notebooks, or downstream pipelines.

  • Snapshotting before risky migrations.

Install

pipx install git+https://github.com/moabe-br-2019/bubble_cli.git
bubble init
bubble scan
bubble pull --all

Requires Python 3.10+ and a Data API key from your app’s Settings → API. Each type you want to sync needs “Expose for Data API” turned on.

Repo: https://github.com/moabe-br-2019/bubble_cli

Looking for feedback on

  • Edge cases in the schema inference — apps with unusual field types, option sets, nested data, etc.

  • Whether the incremental cursor (Modified Date > last_sync_at) holds up well on larger datasets.

  • Anything you’d expect a tool like this to do that it currently doesn’t.

If you give it a try, issues and PRs are very welcome on the repo. Happy to answer questions here too.

2

niice, will try out for sure when we’re going live.

3

Bubble CLI — updates since launch (now at v0.1.8)

Since its initial release, Bubble CLI — the open-source command-line tool that mirrors a Bubble.io app’s data into a local SQLite database via the Data API — has received a steady stream of updates. Here is a summary of what has changed.

Reliability and correctness

  • Fixed a crash when type names contained slashes.
  • Fixed duplicate-column crashes caused by field names that differ only in letter case.
  • Added automatic retry with exponential backoff on 429 (rate limit) and 5xx errors, making long pulls much more resilient.
  • --dry-run now correctly respects --mode incremental.

Performance

  • Tables can now be downloaded in parallel during pull using the -j/–jobs flag, significantly speeding up syncs for apps with many data types.

Updates and release notes

  • The CLI now checks for new versions and can self-update via pipx or pip.
  • After an update, a one-time “What’s new” panel shows the release notes for the versions the user skipped, in their preferred language (English or Portuguese).

AI / MCP integration

  • The tool now ships with a built-in MCP (Model Context Protocol) server, exposing the CLI to AI clients such as Claude. It runs with a single command: python -m bubble_cli.mcp_server.
  • The interactive menu gained an MCP item with a step-by-step setup guide for connecting AI clients.
  • Several MCP fixes landed along the way: a startup hang, output encoding issues, and language selection.

Other improvements

  • The BUBBLE_LANG environment variable now overrides the saved language preference, which is handy for scripts and CI.
  • The project is MIT-licensed, with an expanded installation guide in the README.

Feedback, bug reports and feature requests are welcome — the developer is actively iterating on the tool.