Genai Toolbox

What googleapis/genai-toolbox MCP server does

The googleapis/genai-toolbox MCP server exposes database operations to MCP-compatible clients and applications. It can provide generic tools for exploring data and schemas, or act as a framework for defining narrower production tools around an organization’s own data sources.

The prebuilt mode supplies standard database capabilities without requiring each client integration to implement database-specific logic. Examples include list_tables and execute_sql. A database-specific toolset can be selected, such as the PostgreSQL data toolset, rather than loading every available tool.

The custom configuration mode is intended for controlled workflows. Developers can define sources, tools, toolsets, and prompts in tools.yaml. A tool can reference a configured source, declare typed parameters, and execute a predefined statement. This supports use cases such as name-based record searches, structured queries, and other application-specific database actions.

How it works

An MCP client starts the server and communicates with it using the configured transport. The server then connects to one or more database sources and exposes either prebuilt tools or the resources declared in the YAML configuration.

For prebuilt access, the startup command includes a database selection such as --prebuilt=postgres. The optional /toolset suffix narrows the loaded capabilities. For custom deployments, --config tools.yaml points the server to the configuration file containing source and tool definitions.

Supported sources include Google Cloud databases and services such as AlloyDB, BigQuery, Cloud SQL, Spanner, Firestore, and Knowledge Catalog. Other listed integrations include PostgreSQL, MySQL, MariaDB, SQL Server, Oracle, MongoDB, Redis, Elasticsearch, CockroachDB, ClickHouse, Couchbase, Neo4j, Snowflake, and Trino.

Setup and configuration

The googleapis/genai-toolbox MCP server can be configured in an MCP client’s server configuration, commonly a file such as mcp.json or claude_desktop_config.json. A local npm setup uses the project’s server package and passes a prebuilt database type:

{
  "mcpServers": {
    "toolbox-postgres": {
      "command": "npx",
      "args": ["-y", "@toolbox-sdk/server", "--prebuilt=postgres", "--stdio"]
    }
  }
}

Database connection settings are supplied through the variables required by the selected prebuilt configuration. For custom tools, create a tools.yaml file with source, tool, toolset, and, where needed, prompt resources. The source definition identifies connection details such as database type, host, port, database name, user, and password.

The repository also documents Docker and binary-based execution methods. The project has been renamed from genai-toolbox to mcp-toolbox, although the supplied repository URL reflects the former name.

Tools and capabilities

The googleapis/genai-toolbox MCP server supports:

• Prebuilt generic tools for database exploration and SQL execution.
• Database-specific prebuilt toolsets.
• Custom tools with declared parameters and predefined statements.
• Toolsets that group capabilities for different agents or applications.
• Prompt resources defined in the same configuration model.
• Connection pooling and integrated IAM authentication.
• Metrics and tracing through OpenTelemetry.
• SDK-based integration for Python, JavaScript/TypeScript, Go, and Java applications.

Limitations and notes

The server requires database-specific connection configuration; the excerpt does not provide the exact environment variable names for each source. Available tools and settings vary by database integration and selected toolset. Although the project is open source, access to a connected database still depends on that database’s credentials and authorization. The README names Gemini CLI, Google Antigravity, Claude Code, Codex, and other MCP clients, while the example specifically shows Claude Desktop configuration.