LabscoMCP early access ↗
googleapisMCP SERVER · OFFICIAL PROJECT

MCP Toolbox for Databases

Point an IDE at your database with prebuilt tools, or define your own tools in YAML and give agents only those.

by googleapis ↗Relational SQL Databases✓ Official

LABSCO SUMMARY

One server that is both an instant database connector and the place you lock down what agents may run.

The prebuilt path is a two-minute setup that hands an editor list_tables and execute_sql against a real database, which is where most people start. The tools.yaml path is the one that survives contact with production: you write the statement, its parameters, and the description, and the agent gets that tool rather than an open SQL prompt. Connection pooling, IAM authentication, and OpenTelemetry come with either path.

What it is

Google's open-source MCP server for enterprise databases, serving two purposes from one binary. A --prebuilt flag gives an MCP client generic tools against a database immediately; a tools.yaml file instead defines exactly which queries an agent may run, and nothing else. The repository was renamed from genai-toolbox to mcp-toolbox.

What you get

  • Prebuilt generic tools per database — list_tables, execute_sql and the rest — reachable with npx -y @toolbox-sdk/server --prebuilt=postgres --stdio
  • Coverage across AlloyDB, BigQuery, Cloud SQL for PostgreSQL, MySQL and SQL Server, Spanner, Firestore, and Knowledge Catalog, plus PostgreSQL, MySQL, MariaDB, SQL Server, Oracle, MongoDB, Redis, Elasticsearch, CockroachDB, ClickHouse, Couchbase, Neo4j, Snowflake, and Trino
  • Custom tools in tools.yaml: a source, a parameterised statement, and the description the model reads — the agent gets your query rather than free-run SQL
  • Toolsets that group tools for one agent or application, addressable at /mcp/{toolset_name}
  • Prompts defined alongside the tools in the same configuration file
  • An interactive UI for exercising tools and toolsets, launched with --ui
  • OpenTelemetry traces and metrics, exported with --telemetry-otlp to any OTLP-compatible backend
  • Skills-generate, which turns a toolset into an installable Agent Skill package
  • Client SDKs for Python, JavaScript and TypeScript, Go, and Java that load the same toolsets into ADK, LangChain, LlamaIndex, or Genkit

Requirements

No account for the server itself; each source in your configuration carries its own database credentials. Run it as a released binary, with brew install mcp-toolbox, from the container image us-central1-docker.pkg.dev/database-toolbox/toolbox/toolbox, or with npx @toolbox-sdk/server. The default transport is Streamable HTTP, listening on 127.0.0.1 port 5000 at /mcp; --stdio switches it to stdio for local clients. --config points at your tools.yaml, and configuration reloads dynamically unless you pass --disable-reload. Current release is 1.9.0.

Setup effort

One command — npx @toolbox-sdk/server --config tools.yaml

See installation options