DBX Setup: One 25 MB Client for 100+ Databases

DBX is a free, Apache-2.0 database client written in Rust that opens more than 100 databases, from PostgreSQL, MySQL and SQLite to Redis, MongoDB, DuckDB and ClickHouse, in an installer of about 27 MB. It gained 1,133 GitHub stars today and has about 23,100 in total, and the same app runs as a desktop client, a self-hosted web version in Docker, a CLI and an MCP server your AI coding agent can query. Installing it and running your first query takes about ten minutes.

• One Rust core, four front doors: desktop app (Windows, macOS, Linux), Docker web app on port 4224, a CLI, and an MCP server for Claude Code, Cursor and similar agents.
• No Java runtime and no bundled Chromium. It is built on Tauri 2, so the Windows installer is 26.8 MB and the web edition is a single executable.
• The AI assistant is optional and bring-your-own-model: Claude, OpenAI, Gemini, DeepSeek, or a free local model through Ollama.
• It ships almost daily. v0.6.29 landed on 30 September, the fifth release in four days, so expect small changes between what you read and what you install.

The exact steps, start to finish

1. Check your prerequisites. You need one package manager for the desktop app, Docker if you want the self-hosted web version, and Node.js 18.18 or newer only for the CLI and MCP server. Run whichever apply:

   # Windows (same in PowerShell and cmd.exe)
   winget --version
   # macOS
   brew --version
   # Linux
   flatpak --version
   # Optional: self-hosting and the CLI/MCP extras
   docker --version
   node --version

2. Install the desktop app. Pick the line for your OS. All three come straight from the project README.

   # Windows (PowerShell or cmd.exe)
   winget install t8y2.dbx
   # macOS
   brew install --cask dbx
   # Linux (any distribution)
   flatpak remote-add --if-not-exists flatpark https://dl.flatpark.org/flatpark.flatpakrepo
   flatpak install flatpark com.dbxio.dbx

3. Or self-host the web version instead. This runs the same workbench in your browser at http://localhost:4224, with data kept in a named volume.

   # macOS / Linux
   docker run -d --pull=always --name dbx -p 4224:4224 \
     -v dbx-data:/app/data \
     t8y2/dbx:latest
   # Windows PowerShell or cmd.exe (same command on one line)
   docker run -d --pull=always --name dbx -p 4224:4224 -v dbx-data:/app/data t8y2/dbx:latest

   The first time you open the web version it asks you to set an access password. Do it before anything else.
4. Add your first connection. Click New Connection, search for your database type (for example SQLite or PostgreSQL), click Next, fill in the file path or host, port, user and password, then press Test and Save & Connect. No account, signup or API key is needed for this step.
5. Run a query. Click New Query, type your SQL and press Ctrl+Enter (Cmd+Enter on a Mac). Results open in a grid underneath with a Chart tab beside them.
6. Optional: turn on the AI assistant. Open Settings → AI → Add Config and choose a preset. For a free, local option install Ollama and pick the Ollama preset; otherwise paste an API key from your Claude, OpenAI or Gemini account. Press Test, then save.
7. Optional: let your coding agent query the same connections. Add the MCP server to your agent's .mcp.json, then choose read only, read/write or full access in Settings → MCP.

   {
     "mcpServers": {
       "dbx": { "command": "npx", "args": ["-y", "@dbx-app/mcp-server"] }
     }
   }

8. Optional: query from the terminal. The CLI reads the same saved connections as the app. Replace local with the name you gave your connection.

   npm install -g @dbx-app/cli
   dbx agent setup
   dbx connections list --json
   dbx query local "select 1" --json

What is DBX and why is it trending?

DBX is a database workbench in the tradition of DBeaver, TablePlus and Navicat: a schema tree on the left, SQL tabs in the middle, a data grid for results. What sets it apart is weight and scope. The desktop build is a Tauri 2 app with a Vue 3 front end and a Rust back end, so it does not drag in a Java runtime or a whole browser engine. The core speaks to relational databases, the Chinese enterprise databases (Dameng, KingbaseES, openGauss and others) that most Western tools ignore, document and key-value stores, vector databases such as Qdrant and Milvus, and message queues including Kafka, RabbitMQ and MQTT. Anything left over can go through an optional JDBC plugin.

RelatedUniver Setup: Embed an Excel-Like Spreadsheet in Your Web App

The feature list is long for a project that only started in April 2026: a CodeMirror 6 editor with schema-aware autocomplete, a virtual-scrolled grid with inline editing that previews the SQL before it commits, an ER diagram, schema diff, data compare, visual explain plans, column-level lineage, CSV and Excel import, and database-to-database transfer. You can also import saved connection profiles from DBeaver or Navicat, which removes most of the switching cost.

The reason it keeps climbing GitHub Trending is the AI angle. DBX treats agents as first-class users: an in-app assistant with Ask and Agent modes, a separately distributed MCP server, and a CLI that installs its own agent skill with dbx agent setup. For anyone already working in Claude Code or Cursor, "my agent can read my database through the connections I already saved, with a central read-only switch" is an easy sell. The project also releases at a relentless pace, which keeps it on the trending page day after day.

How do you install DBX on Windows?

The quickest route is winget install t8y2.dbx, which pulls the current 0.6.29 package. Scoop users can run scoop bucket add dbx https://github.com/t8y2/scoop-bucket followed by scoop install dbx. Both lines are identical in PowerShell and cmd.exe. If you prefer a manual download, the Releases page has a 26.8 MB DBX_0.6.29_x64-setup.exe, a 35.0 MB MSI, a 35.4 MB portable zip, and a separate build for Windows 7 and Server 2012 R2.

For this guide we tested the web edition on a Windows 10 laptop: the 33.3 MB dbx-web zip for Windows unpacks to a single dbx-web.exe plus a start.bat. Run the batch file and it serves the full workbench at http://127.0.0.1:4224 with no Docker involved. We set the access password, pointed a new SQLite connection at a sample shop database, opened the orders table in the grid, and ran a join across two tables that came back in 6 ms. The native CLI (version 0.4.102, a 12.8 MB zip) then read the same data directory and returned the same connection and query results in the terminal.

How do you install it on macOS, Linux or a server?

On a Mac, brew install --cask dbx installs it and brew upgrade --cask dbx keeps it current; the Apple Silicon DMG is 36.4 MB if you would rather download it. On Linux the documented path is Flatpak from FlatPark, which works on any distribution and updates through the usual flatpak update. The Releases page also carries .deb and .rpm packages and an AppImage, which needs chmod +x before its first run.

For a shared team instance, use the Docker command from the checklist or the Compose file the project ships:

docker compose -f deploy/docker-compose.release.yml up -d

The official docs recommend setting -e DBX_PASSWORD=your-password on production containers, and they document DBX_PUBLIC_BASE_PATH for serving it under a sub-path such as /dbx behind a reverse proxy. Images are multi-arch, amd64 and arm64.

RelatedOpenWork Setup: Open-Source Claude Cowork in 15 Minutes

How does DBX compare with DBeaver, TablePlus and Beekeeper Studio?

What are the gotchas before you rely on it?

The web edition listens on every network interface. When we ran dbx-web.exe its log reported the server starting on 0.0.0.0:4224, so on a laptop on shared Wi-Fi anyone on that network can reach the login page. The first-run password screen is your protection, and the docs also document a DBX_DISABLE_PASSWORD switch that you should never flip on a machine that other people can reach.

The CLI is not a full replacement for the app. dbx doctor splits databases into "direct query" types (PostgreSQL, MySQL, SQLite and a handful of others) and a much longer list of "bridge-required" types, including Redis, MongoDB, DuckDB, SQL Server and Oracle, that only work while the desktop app is running as a bridge. The MCP server is also a separate install: putting DBX on your machine does not put @dbx-app/mcp-server there.

Guard your data directory. Saved connection passwords are encrypted, using the OS credential store on desktop and a key file at ${DBX_DATA_DIR}/.dbx/secret.key for Docker and the web binary. Back that key up together with dbx.db, or your saved credentials become unreadable after a restore. Finally, the GitHub release notes are written in Chinese. The English docs site is solid, but if you track changes release by release you will want a translator open.

Our take

DBX is the most interesting new database client we have tried this year, and not only because it is small. It gets the modern workflow right: one set of saved, encrypted connections that a person, a script and an AI agent can all use, with read-only and production guards that the agent cannot widen. The web edition running from a single Windows executable was a genuinely pleasant surprise. The caveats are the ones you would expect from a project a few months old that ships every day. Use it as your daily driver for local and staging databases now, keep DBeaver around for the odd driver it lacks, and give it a read-only account before you let it, or your agent, anywhere near production.

Original analysis by GenZTech, based on the DBX repository and documentation and our own test of the v0.6.29 web edition and v0.4.102 CLI on Windows 10.