DBHub
3.6k stars · Open source · TypeScript
One small server for Postgres, MySQL, MariaDB, SQL Server, Oracle and SQLite, with only two tools by default to keep context lean, and read-only mode, row limits and query timeouts built in.
The shortlist
Connect an AI agent to your database
An MCP server is the standard way to let an agent inspect schemas and run queries against a real database. The choice is mostly about which databases you run and how much the server should do beyond querying.
Postgres MCP Pro
3.3k stars · Open source · Python
For Postgres it goes beyond queries: database health checks, index tuning across candidate indexes and EXPLAIN plans with hypothetical indexes, with restricted and unrestricted access modes.
MCP Toolbox for Databases
16k stars · Open source · Go
Google's open-source server covers a long list of databases, from Postgres and BigQuery to Spanner, with prebuilt tools for exploration and a framework for governed custom tools once agents move to production.
Before you commit
Start read-only and against a copy. A model with write access will eventually run a statement you did not intend, so use DBHub's read-only mode or Postgres MCP Pro's restricted mode, point the server at a replica or development database, and never hand it credentials that can drop tables. If your data lives in Supabase or MongoDB, both vendors publish their own MCP servers, listed in the index.
More decisions
- Automate browser tasks
- Connect your coding agent to GitHub
- Research the web with citations
- Build a production AI agent
- Build a customer-support agent
- Build a voice receptionist
- Run an agent locally
- Automate work without code
- Feed your agent clean web data
- Give your agent live docs & search
- Pick a coding agent
- Build a knowledge assistant over your docs
- Orchestrate a team of agents
- Turn Figma designs into code
- Give your AI agent long-term memory
- Use an AI agent inside your game engine