MCP Toolbox for Databases

Install

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

This is third-party code. Review the repository files before installing.

What it does

Toolbox connects an agent to a database in two ways. With --prebuilt you get standard tools such as list_tables and execute_sql for the chosen database, handy for exploring schemas and writing code. In the second mode tools.yaml defines sources, specific queries, toolsets, prompts and resources so the agent gets only approved operations. The server handles connection pooling, IAM auth and OpenTelemetry tracing, and there are SDKs for ADK, LangChain and LlamaIndex.

Who it is for. For developers and analysts who need agent access to databases, including Google Cloud ones.

Example request

List the tables in the database and write a query that counts orders by month

Limitations

Database connection settings are passed via environment variables; the set depends on the database and is listed in the prebuilt configs reference. Prebuilt tools include execute_sql, so use a read-only account or custom tools on production databases. Local databases need no VPN; Google Cloud database availability from Russia is not guaranteed. The authors consider npx less reliable than the binary or container.

How to disable. Delete the toolbox block from the client's MCP config and stop the toolbox process or container.

MCP

Transport

stdio, http

Authentication

not required

Security check

• Can run arbitrary SQL against the connected database
• Handles database credentials

README in short

The README explains the two modes: prebuilt tools and the custom tools framework. It gives an MCP config example for Postgres and walks through tools.yaml sections. It covers installation via binary, container, Homebrew, go install and Gemini CLI extensions, plus SDK integration. Apache-2.0 licensed.

Related