DBHub MCP server

DBHub is an open-source MCP server by Bytebase that gives AI agents SQL access to PostgreSQL, MySQL, MariaDB, SQL Server, Oracle and SQLite through one server with two default tools.

Checked by CookiesForAI on : repository facts (stars, last push, archived) re-read from GitHub; listing in the MCP Registry checked.

Written by

Connect DBHub

Claude Code, as a plugin (prompts for the connection string and stores it securely)

Agent chat
/plugin marketplace add bytebase/dbhub

Then install the plugin

Agent chat
/plugin install dbhub@dbhub

Codex, trying the built-in demo database first

Terminal
codex mcp add dbhub -- npx @bytebase/dbhub@latest --transport stdio --demo

Run as an HTTP server for any client (add your own --dsn instead of --demo)

Terminal
npx @bytebase/dbhub@latest --transport http --port 8080 --demo

Pass your database with --dsn, for example a postgres:// URI with user and password, or keep several connections in a dbhub.toml file and use --config. From Docker, use host.docker.internal instead of localhost to reach a database on your machine.

Or paste this into your coding agent: Connect the DBHub MCP server to my coding agent using its documented setup: https://github.com/bytebase/dbhub. Checked by CookiesForAI on : repository facts (stars, last push, archived) re-read from GitHub; listing in the MCP Registry checked.

Your first question with this server installed

371tokens sent to the model with your first question: the tool list plus one tool answer

DBHubTool list 274One answer 972 tools listed without a key

Typical serverFirst question 2,022 (median of 10 servers)

DBHub's first question takes 371 tokens: its tool list plus one answer. That is about 18% of the median first question of the 10 servers we measured with both numbers on (2,022 tokens).

Measured without a key or sign-in. Assistants that load every tool send this list to the model with each question; some load tools only when a task needs them. The one answer is from execute_sql, a fixed read-only call. Our page names 4 of its tools, so set up as its maker documents, it can list more than we measured. Tokens are counted with OpenAI's o200k tokenizer; other models count a little differently.

Measured . How we measure.

What the DBHub MCP server does

DBHub is a database MCP server from Bytebase, the company behind a database DevOps platform. One DBHub process connects an AI agent to PostgreSQL, MySQL, MariaDB, SQL Server, Oracle or SQLite, and a TOML file lets it hold several connections at once, for example a local dev database and a read-only replica. It supports SSH tunnels and SSL for databases behind a bastion host.

DBHub keeps its default tool list to two: execute_sql runs queries with transaction support, and search_objects explores schemas, tables, columns, indexes and procedures a step at a time. Bytebase measures this at about 1,400 tokens of context, against roughly 19,000 for some alternatives. Two opt-in tools show a query plan and report connection pool health, and you can define your own parameterized SQL tools in the config file. A built-in web workbench runs the same tools without an MCP client.

Guardrails include a read-only mode, a row limit and a query timeout. DBHub runs through npx (Node.js 22.5 or newer), Docker, an MCP Bundle file or a Claude Code plugin; the bundle and plugin are read-only by design. Claude Code's own docs use DBHub as the example for querying PostgreSQL. The repo had 3,589 GitHub stars on .

Tools it gives your agent

  • execute_sql
  • search_objects
  • explain_sql
  • health_check

When to use DBHub

  • You work with more than one database engine and want a single MCP server for all of them.
  • You want the agent to explore an unfamiliar schema without filling its context with tool definitions.
  • You want to give non-technical teammates read-only views of data through Claude Desktop or VS Code.
  • You need to reach a database behind an SSH bastion host.

When to pick something else

  • Deep Postgres tuning such as index recommendations and health reports; Postgres MCP Pro specializes in that.
  • Write access on production without care: unless read-only mode is on, execute_sql can change data.

What DBHub needs

  • Node.js 22.5 or newer, or Docker
  • A database connection string (DSN)

What the DBHub MCP server costs

Free and open source under MIT and run on your own machine, with no paid tier; Bytebase's enterprise product is separate.

Questions people ask

Is DBHub free?

Yes. DBHub is open source under MIT and runs on your own machine with no paid tier.

Which databases does DBHub support?

PostgreSQL, MySQL, MariaDB, SQL Server, Oracle and SQLite (README read ). SQLite needs no extra driver because DBHub uses Node's built-in module.