MCP Toolbox for Databases is an open-source Model Context Protocol (MCP) server that connects AI agents, IDEs, and applications to databases. It began as Gen AI Toolbox for Databases, before MCP existed, and was renamed when it adopted the protocol.
It serves two purposes. At build time it is a ready-to-use MCP server: a prebuilt flag per database gives Gemini CLI, Google Antigravity, Claude Code, Codex, or any other MCP client a standard set of tools (list tables, execute SQL, and similar) to query data in plain language, explore and manage schemas, and generate database-aware code. At run time it is a framework for custom tools: a YAML manifest declares sources, tools, toolsets, prompts, and resources, so an agent in production only ever calls predefined structured queries, semantic search, or NL2SQL tools against the databases it is allowed to reach.
Supported databases include AlloyDB, BigQuery, Cloud SQL (PostgreSQL, MySQL, SQL Server), Spanner, Firestore, and Knowledge Catalog on Google Cloud, plus PostgreSQL, MySQL, MariaDB, SQL Server, Oracle, MongoDB, Redis, Elasticsearch, CockroachDB, ClickHouse, Couchbase, Neo4j, Snowflake, and Trino. The server handles connection pooling, IAM-integrated authentication, and OpenTelemetry metrics and traces that export to any OTLP backend.
Client SDKs for Python, JavaScript/TypeScript, Go, and Java load a toolset from the server and hand the tools to LangChain, LangGraph, LlamaIndex, Agent Development Kit (ADK), Genkit, OpenAI Go, or a custom agent in a few lines. The server runs as an npm package, a binary, or a container image, and a Gemini CLI extension installs prebuilt tools for specific Google Cloud databases. An interactive UI tests tools and toolsets, and a skills-generate command converts a toolset into an Agent Skill package that a Gemini CLI can install.
Features
- Prebuilt database tools: a per-database flag exposes standard tools such as list tables and execute SQL to any MCP client, optionally narrowed to one toolset
- Custom tools framework: a YAML manifest defines sources, parameterized SQL and semantic-search tools, toolsets, prompts, and read-only resources that an agent can call
- Database coverage: AlloyDB, BigQuery, Cloud SQL, Spanner, Firestore, PostgreSQL, MySQL, MariaDB, SQL Server, Oracle, MongoDB, Redis, Elasticsearch, CockroachDB, ClickHouse, Couchbase, Neo4j, Snowflake, and Trino
- Client SDKs: Python, JavaScript/TypeScript, Go, and Java clients that load toolsets into LangChain, LangGraph, LlamaIndex, ADK, Genkit, OpenAI Go, or custom agents
- Authentication and pooling: integrated auth (IAM) and connection pooling handled by the server rather than each agent
- Observability: OpenTelemetry metrics and traces exported to any OTLP-compatible backend
- MCP client access: HTTP or stdio transport, with per-toolset endpoints and a one-click install through the Google Antigravity MCP Store
- Toolbox UI: an interactive page for testing tools and toolsets, including authorized parameters
- Agent Skills: a toolset can be packaged as an Agent Skill that installs into Gemini CLI, and the repository ships ready-made skills
- Semantic versioning: a public API covering the server CLI, configuration manifests, prebuilt toolsets, and the client SDKs
