MCP Toolbox for Databases

MCP Toolbox for Databases is Google's open-source MCP server for databases. Connect your agent to a database and it can explore the schema, write and run SQL, and check on the database's health, so you can talk to your data and generate database-aware code without writing boilerplate. What you can do with it - Ask questions about your data in plain English and let your agent run the SQL - Explore schemas: tables, views, indexes, triggers and stored procedures - Find long-running queries, locks and query plans - Check database health, such as bloated tables and invalid indexes - Generate code that matches your real schema Databases it supports - Google Cloud: AlloyDB, BigQuery, Cloud SQL (PostgreSQL, MySQL, SQL Server), Spanner, Firestore and Knowledge Catalog - Others: PostgreSQL, MySQL, MariaDB, SQL Server, Oracle, MongoDB, Redis, Elasticsearch, CockroachDB, ClickHouse, Couchbase, Neo4j, Snowflake, Trino and more How it connects It runs on your machine. Add it to your agent with npx @toolbox-sdk/server and a --prebuilt flag for your database (for example --prebuilt=postgres), then set environment variables with your connection details. You can also run it with Docker or a binary, install it from the Google Antigravity MCP Store, or add it as a Gemini CLI extension. Choose what it can do Load a smaller toolset with --prebuilt=<database>/<toolset>, such as postgres/data for just the SQL tools. Teams can also define their own tools in a tools.yaml file, with predefined queries, so an agent only runs what you allow. Good to know The prebuilt execute_sql tool runs whatever SQL your agent writes, so connect with a database user that has only the access you want to give it.