Database clients are traditionally heavy: DBeaver needs a Java runtime, many rivals embed a browser engine, and older tools such as Navicat are paid. The open-source project DBX takes the opposite route and ships a single binary of about 25 megabytes for macOS, Windows and Linux — built with Tauri 2, a Rust backend and a Vue 3 front end, with no bundled Chromium, no JRE and no Python environment.
The real claim lies in driver coverage. DBX natively supports, by its own catalogue, more than ninety systems: MySQL, PostgreSQL, SQLite, Redis, MongoDB, DuckDB, ClickHouse, SQL Server, Oracle, Elasticsearch, MariaDB, TiDB, CockroachDB, Redshift, Doris and StarRocks, alongside Chinese engines such as Dameng, OceanBase, openGauss and KingbaseES, plus vector and search services including Qdrant, Milvus, Weaviate and Meilisearch. Additional systems — Snowflake, Trino, Hive, DB2, Neo4j, Cassandra, BigQuery, Databricks, SAP HANA, Teradata — run through agent-based JDBC profiles. Message queues such as Kafka, Pulsar and RocketMQ can be administered too. Users arriving from DBeaver or Navicat can import their connection profiles.
Two features tie the tool to the current wave of developer tooling. The first is an AI SQL assistant that generates SQL from a description, explains, optimises and repairs queries, and routes AI-generated statements through built-in safety checks before execution; backends include Claude, OpenAI, local models via Ollama or any OpenAI-compatible endpoint. The second is a separate Rust MCP server, distributed independently of the desktop app and wired into `.mcp.json` through `npx -y @dbx-app/mcp-server`. Coding agents such as Claude Code, Cursor and Codex can then query databases over connections the user already configured, browse tables and run SQL. Permissions are governed centrally by a connection allowlist and three modes — Read only, Data read/write and Full access, internally `read_only`, `safe_write` and `high_risk_write` — and a legacy environment flag can never widen a saved policy.
For teams there is a Docker build that serves a web interface on port 4224, with multi-arch images. The treatment of secrets is notably careful: connection, plugin, AI and tunnel credentials are encrypted into `dbx.db`, with the key held in the platform credential store on the desktop and in `${DBX_DATA_DIR}/.dbx/secret.key` under Docker or the web build, replaceable by `DBX_SECRET_KEY_FILE` or `DBX_SECRET_KEY` from a secret manager. Installations that still carry plaintext credentials are pushed through a migration wizard with backup and verification on startup — a detail many comparable tools still lack. For a 25 MB utility, that is an unusually complete security model.
Limits remain: the project is Apache-2.0 and funded by sponsors and donations; its list of new native drivers is growing faster than its documentation; and anyone needing an exotic database from the agent-based list has to supply JDBC drivers. But for developers who maintain many systems and simultaneously want agents to have governed access to data, DBX is one of the clearest expressions of where this tool market is tilting: small binaries, local execution, MCP from the start.
