t8y2 / dbx

Database client · desktop, Docker, CLI, MCP

100+ databases.
25 megabytes.

MySQL, PostgreSQL, SQLite, Redis, MongoDB, DuckDB, ClickHouse, SQL Server, Oracle, Elasticsearch and dozens more — plus agent-driven profiles for the long tail. All in a single ~25 MB app. No bundled Chromium.

DesktopDockerCLIWeb Built-in AI assistantMCP serverPlugin store

Size against reach

The two numbers the project leads with, drawn to the same scale.

the download~25 MB
what it speaks100+ engines

relationaldocumentkey-value searchtime-seriesqueue & config

Bar lengths are illustrative and not to a shared unit — the point is the direction, and both figures are the project’s own. What is doing the work is the absence of a bundled browser engine.

When an agent connects

three modes, named before you wire it up
Read only

It can look

The agent queries the connections you configured in DBX and returns rows. Nothing changes.

read_only
Data read / write

It can edit rows

Data operations are permitted. Schema and structure stay where you left them.

safe_write
Full access

It can do the rest

The project names this one high risk in its own machine-readable value, which is the honest way to label a mode.

high_risk_write

The default cannot be widened by accident. The project states that an existing DBX_MCP_ALLOW_WRITES=0 stays a read-only restriction until a central MCP policy is saved — and that it can never enable writes or override a saved policy.

The MCP server ships separately from the desktop app, so installing DBX does not put an agent-reachable database bridge on your machine. That is a packaging decision, and it is the kind that only shows up in a README.