Google open-sources MCP Toolbox for Databases: an MCP server for databases

MCP Toolbox for Databases is an open-source Model Context Protocol server that connects AI clients directly to enterprise databases. It works two ways: prebuilt tools, where --prebuilt=<database> lets clients like Gemini CLI, Google Antigravity, Claude Code, and Codex query tables, run SQL, and explore schemas; and a custom-tool framework, where tools.yaml defines source, tool, toolset, prompt, and resource, paired with restricted access, structured queries, and semantic search for production-grade tools. The repo has been renamed from genai-toolbox to mcp-toolbox. Supported databases include Google Cloud's AlloyDB, BigQuery, Cloud SQL, Spanner, and Firestore, plus PostgreSQL, MySQL, SQL Server, Oracle, MongoDB, Redis, Neo4j, and Snowflake. Tools have built-in connection pooling, IAM auth, and OpenTelemetry-based observability, and can integrate with ADK, LangChain, and LlamaIndex in under ten lines of code. Install via binary, container image, Homebrew, or source; current version is 1.14.0. For teams building production agents, this eliminates repetitive database connection management and observability work, but you still need to enforce tools.yaml security boundaries and custom SQL permissions yourself; prebuilt tools are better suited for letting IDE AI assistants query data and generate schema-related code.