dbhub

DBHub is a lightweight MCP (Model‑Controlled‑Prompt) server that lets AI‑enabled coding assistants talk directly to several popular relational databases. It runs with only two built‑in tools, keeping token usage low, and can be extended with optional extras. The service is hosted at https://dbhub.ai/.

The server acts as a gateway between databases such as PostgreSQL, MySQL, SQL Server, Oracle, MariaDB, and SQLite and any MCP‑compatible client. You configure connections in a TOML file, then AI agents call the /mcp endpoint to execute the provided tools. The default tools are execute_sql (for running queries with transaction support and safety controls) and search_objects (for exploring schemas, tables, columns, indexes, and procedures). Optional tools include explain_sql (to retrieve an execution plan without running the query) and health_check (to report connection‑pool state). You can also define custom, parameterized SQL operations that the AI can reuse.

Typical use cases include local development work such as schema exploration, query validation, and data debugging when using editors like Claude Desktop, Claude Code, VS Code, or Cursor. Non‑technical staff can be given read‑only views through the same clients, while a single DBHub process can replace multiple MCP servers in a multi‑database environment. Production troubleshooting is supported by read‑only mode, row limits, and query timeouts that act as guardrails against runaway queries.

DBHub provides a web‑based Workbench where humans can run the same database tools and see request traces without needing an MCP client. Installation options include Docker, NPM, the MCP bundle, or a Claude Code plugin, and the server can be secured with SSH tunneling or SSL/TLS encryption. Because the core package loads at roughly 1.4 k tokens, it leaves most of the context window available for the AI's actual work, which is considerably lower than many alternatives.

Key features

• Token‑efficient core (≈1.4 k tokens) with only two default MCP tools
• Supports PostgreSQL, MySQL, MariaDB, SQL Server, Oracle, and SQLite through a single interface
• Read‑only mode, row‑limiting, and query timeout guardrails
• Optional tools: explain_sql and health_check plus custom reusable SQL operations
• Web Workbench for direct browser‑based tool execution and trace viewing
• Secure connections via SSH tunneling and SSL/TLS