Submit Log in

MCP Toolbox for Databases

by Google

3 views

v1.8.0

Apache-2.0

Repository Docs

Featured Databases & Data Cloud & Infrastructure

databasessqlbigquerypostgresgoogle-cloud

Summary

Google's open-source MCP server for enterprise databases — one binary that fronts AlloyDB, BigQuery, Spanner, Postgres, MySQL, Oracle, MongoDB, Snowflake and more.

Features

One MCP server across AlloyDB, BigQuery, Spanner, Postgres, MySQL, Oracle, MongoDB, Snowflake and more
Custom tools declared in YAML, so agents get scoped queries instead of arbitrary SQL
Automatic connection pooling and integrated authentication
Built-in OpenTelemetry metrics and tracing
Interactive --ui console for testing tools before wiring an agent

Installation

Set up this MCP server in your favorite AI agent — copy a ready-made configuration below.

Any MCP-compatible agent

Most agents (Claude, Cursor, Windsurf, VS Code, and more) read a standard mcpServers configuration.

Open your agent's MCP configuration file.
Merge the snippet below into it.
Restart the agent — the "MCP Toolbox for Databases" tools will be available.

{
  "mcpServers": {
    "mcp-toolbox-databases": {
      "command": "toolbox",
      "args": [
        "--tools-file",
        "tools.yaml",
        "--stdio"
      ]
    }
  }
}

Claude Code

Add the server with a single CLI command.

claude mcp add mcp-toolbox-databases -- toolbox --tools-file tools.yaml --stdio

Claude Desktop

Add the server to claude_desktop_config.json (Settings → Developer → Edit Config).

{
  "mcpServers": {
    "mcp-toolbox-databases": {
      "command": "toolbox",
      "args": [
        "--tools-file",
        "tools.yaml",
        "--stdio"
      ]
    }
  }
}

Install the binary with Homebrew, then run it with --tools-file tools.yaml.

{
  "command": "brew",
  "args": [
    "install",
    "mcp-toolbox"
  ]
}

Description

Giving an agent database access usually means one MCP server per engine, each with its own auth story and its own idea of what a "query" tool looks like. MCP Toolbox is Google's answer to that: a single Go binary, configured by a YAML file, that fronts a long list of databases behind one MCP endpoint.

It works two ways. Prebuilt toolsets get an agent querying an existing database immediately. The custom-tools framework is the more interesting half — you declare each tool in tools.yaml with its own source, parameters and SQL, so the agent gets a narrow, reviewable surface ("look up an order by id") instead of arbitrary SQL execution against production. Tools group into toolsets, and prompts can be declared alongside them.

Sources span Google Cloud — AlloyDB, BigQuery, Cloud SQL for PostgreSQL, MySQL and SQL Server, Spanner, Firestore, Knowledge Catalog — and a wide set of everything else: PostgreSQL, MySQL, MariaDB, SQL Server, Oracle, MongoDB, Redis, Elasticsearch, CockroachDB, ClickHouse, Couchbase, Neo4j, Snowflake and Trino.

Operationally it behaves like a server rather than a script: automatic connection pooling, integrated authentication so access is scoped rather than shared, and built-in OpenTelemetry metrics and tracing. A --ui flag opens an interactive console for exercising tools before an agent ever calls them, and client SDKs exist for Python, JavaScript/TypeScript, Go and Java for use outside MCP.

Apache-2.0 licensed. Install via Homebrew (brew install mcp-toolbox), a released binary, npm, or the published container image.

Related MCP Servers

MCP: Weaviate MCP Server

by Weaviate

New

Weaviate's MCP server, built into the database itself — hybrid search, schema inspection, tenant listing and batch upserts at /v1/mcp, scoped by RBAC.

Databases & Data