DBX: A 15MB Database Client With Built-In AI and MCP Support

DBX is a 15MB, open-source database client that supports 40+ databases — with a built-in AI SQL assistant and native MCP server support for AI coding agents. It ships as a single binary with zero runtime dependencies: no Java, no Python, no bundled Chromium. Available on macOS, Windows, Linux, Docker, and as a web app.

If you’ve ever waited for DBeaver to launch — watching Java churn through its startup ritual — you know the pain. Database clients have been bloated for years. DBeaver bundles a JRE. DataGrip needs a JetBrains subscription. TablePlus is slick but macOS-only (with limited Windows/Linux support). And none of them were built for the AI-native workflow developers are adopting in 2026.

DBX changes this equation.

What Makes DBX Different?

The size alone is remarkable. DBeaver’s installer is 150MB+. DBX is 15MB — reminiscent of the Portable AI USB philosophy where less is more. It’s built with Tauri 2 (Rust backend, Vue 3 frontend), which explains the tiny footprint and native performance.

But size is table stakes. What makes DBX genuinely interesting for the AI/developer crowd is two features: the AI SQL assistant and MCP integration.

How Does the AI SQL Assistant Work?

DBX has a built-in AI assistant that works with Claude, OpenAI, or local models via Ollama. Highlight a table, describe what you want in plain language, and get SQL back. It can explain queries, optimize SQL, fix errors, and — critically — runs AI-generated SQL through built-in safety checks before execution.

This isn’t a plugin or an afterthought. It’s baked into the editor experience. You don’t copy-paste between ChatGPT and your database client. If you’ve seen tools like QueryWeaver tackling the text-to-SQL problem, DBX bundles that capability directly into a full-featured database client.

The Ollama support is worth highlighting: if you’re working with sensitive data and can’t send schemas to cloud APIs, you can run everything locally. That’s a real consideration for teams with compliance requirements.

Why Does DBX’s MCP Support Matter?

Here’s where DBX gets ahead of every other database client on the market.

DBX ships an MCP server — meaning AI coding agents like Claude Code, Cursor, and Windsurf can query your databases directly through connections you’ve already configured in DBX. We’ve covered the MCP database landscape before — from Google’s managed servers vs MindsDB’s federated approach to Universal DB MCP’s natural language querying — but DBX takes a different angle: it’s a full GUI database client that also speaks MCP.

Setup is dead simple:

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

Drop that in your .mcp.json and your AI coding agent can list connections, browse table schemas, execute queries, and even open tables in DBX’s UI. Your agent doesn’t need database credentials hardcoded — it uses the connections you’ve already set up and secured in DBX.

This is a big deal. The gap between “AI that writes code” and “AI that understands your data” has been one of the friction points in agentic workflows — something OpenViking from ByteDance has also been tackling from the context database angle. DBX bridges it at the tool level.

There’s also a CLI package (@dbx-app/cli) for terminal and script workflows — useful for CI pipelines or Codex-style agents that prefer command-line interfaces.

What Databases Does DBX Support?

DBX supports 40+ databases in a single ~15MB app:

• Relational: MySQL, PostgreSQL, SQLite, MariaDB, SQL Server, Oracle, CockroachDB, TiDB, OceanBase
• Analytics: DuckDB, ClickHouse, Doris, StarRocks, Redshift
• NoSQL: Redis, MongoDB (with Atlas support), Elasticsearch
• Via JDBC plugins: Snowflake, Trino, Hive, BigQuery, Neo4j, Cassandra, DB2

Everything Else Worth Knowing

Beyond the AI angle, DBX is a capable database client:

• Query editor with CodeMirror 6, SQL autocomplete, formatting, diagnostics, and 9 themes
• Virtual-scrolled data grid that handles large result sets with inline editing
• Schema tools including ER diagrams, schema diff, explain plans, and field lineage
• Data operations — import CSV/Excel, transfer between databases, drag-and-drop Parquet/JSON/CSV preview (powered by DuckDB)
• Specialized browsers for Redis (key patterns, TTL editing, all data types) and MongoDB (document CRUD)
• Deployment flexibility — native desktop, Docker self-hosting (docker run -d -p 4224:4224 t8y2/dbx), and web version

How Does DBX Compare?

	DBX	DBeaver	TablePlus
Size	~15MB	~150MB+	~50MB
Runtime	None	Java JRE	None
AI Assistant	Built-in	Plugin	No
MCP Support	Native	No	No
Platforms	macOS/Win/Linux/Docker/Web	macOS/Win/Linux	macOS (primary)
Price	Free (AGPL-3.0)	Free/Paid	Paid

Bottom Line

DBX is what happens when someone builds a database client in 2026 instead of maintaining one from 2010. It’s small, fast, AI-native, and open source. The MCP integration alone makes it worth trying if you’re using AI coding agents — and the 15MB install means there’s essentially no cost to giving it a shot.

Install via Homebrew (brew install --cask t8y2/tap/dbx), Scoop on Windows, or grab a release from GitHub.