AI Native Landscape

mcp-toolbox

Tracked

An open source Model Context Protocol (MCP) server that connects AI agents, IDEs, and applications directly to enterprise databases, with prebuilt generic tools plus a framework for building custom, secure tools. Formerly Gen AI Toolbox for Databases.

AuthorgoogleapisOpen Sourced2024-06-07Last CommitUnknown

Overview

MCP Toolbox for Databases, formerly Gen AI Toolbox (genai-toolbox), is an open source MCP server that connects AI agents, IDEs, and applications directly to enterprise databases. It serves a dual purpose: ready-to-use prebuilt tools that let MCP clients like Gemini CLI, Claude Code, and Codex talk to your data immediately, and a framework for building specialized, production-grade tools with structured queries, semantic search, and NL2SQL behind restricted access.

Key Features

  • Prebuilt generic tools (list_tables, execute_sql, and more) enabled instantly with --prebuilt=<database>, including click-to-install via the Google Antigravity MCP Store
  • Custom tools framework configured through tools.yaml: predefined structured queries, semantic search, and NL2SQL capabilities with restricted access
  • Broad engine support — AlloyDB, BigQuery, Cloud SQL, Spanner, Firestore, Knowledge Catalog, PostgreSQL, MySQL, MariaDB, SQL Server, Oracle, MongoDB, Redis, Elasticsearch, CockroachDB, ClickHouse, Couchbase, Neo4j, Snowflake, Trino, and more
  • Client SDKs in Python, JS/TS, Go, and Java that integrate tools into ADK, LangChain, LlamaIndex, or custom agents in under 10 lines of code
  • Connection pooling, integrated authentication (IAM), and end-to-end OpenTelemetry observability out of the box

Use Cases

  • Connecting an IDE or CLI MCP client to a database to query data in plain English, explore schemas, and generate database-aware code
  • Building production agent tools with your own predefined query logic instead of exposing raw database access
  • Giving LangChain, LlamaIndex, or Agent Development Kit applications safe, typed database tools

Technical Details

  • Tools are declared declaratively in tools.yaml (multiple files via --config): sources, tools, and toolsets are defined separately and composed, so a tool binds parameters to a predefined statement rather than accepting arbitrary SQL
  • Prebuilt mode loads per-engine standard toolsets over stdio — --prebuilt=<database> or a subset like --prebuilt=postgres/data — and is also distributed as the npm package @toolbox-sdk/server for npx-based MCP client configuration
  • The server handles connection pooling and IAM-based authentication, and emits metrics and traces through built-in OpenTelemetry support
  • Four official SDKs (Python toolbox-core, JS/TS @toolbox-sdk/core, Go, Java) load tool definitions from the running server and expose them as native methods for agent frameworks