MCP Toolbox for Databases

What MCP Toolbox for Databases does

MCP Toolbox for Databases exposes database operations to AI agents through the Model Context Protocol. Its ready-to-use configurations provide generic tools for exploring schemas, querying data, and generating code in MCP-compatible clients and IDEs. Common prebuilt tools include list_tables and execute_sql.

The server supports Google Cloud data services such as AlloyDB, BigQuery, Cloud SQL, Spanner, Firestore, and Knowledge Catalog. It also supports PostgreSQL, MySQL, MariaDB, SQL Server, Oracle, MongoDB, Redis, Elasticsearch, CockroachDB, ClickHouse, Couchbase, Neo4j, Snowflake, Trino, and other databases.

It also serves as a framework for application-specific tools. Developers can define restricted access, structured queries, semantic search, and NL2SQL behavior instead of exposing only generic database operations.

How it works

For a quick setup, start the server with a --prebuilt=<database> option. This loads the standard tools for the selected database. A toolset can be selected with a path such as --prebuilt=postgres/data when only a particular group of tools is needed.

For custom behavior, Toolbox reads a tools.yaml configuration. Sources describe the databases available to the server, tools define agent actions and their parameters, toolsets group tools for different agents or applications, and prompts define reusable interactions with language models. A SQL tool can, for example, accept a string parameter and run a parameterized query against a configured PostgreSQL source.

MCP clients connect to the server through its MCP configuration. Toolbox also provides Python, JavaScript/TypeScript, Go, and Java SDKs for integrating its tools into applications and agent frameworks such as ADK, LangChain, and LlamaIndex.

Setup and configuration

A client configuration can launch the prebuilt server with npx, the @toolbox-sdk/server package, a selected database, and the --stdio transport. The database-specific connection settings are supplied according to the selected prebuilt configuration.

Custom deployments use one or more YAML files. Pass the desired file with --config tools.yaml. The configuration can contain source definitions with connection details, tool definitions with SQL or other actions, toolsets, and prompts. Toolbox can also be run using Docker or compiled binaries.

Tools and capabilities

• Prebuilt database tools for schema inspection and SQL execution
• Database-specific toolsets that can be loaded independently
• Custom tools backed by predefined queries and application logic
• Structured query, restricted-access, semantic-search, and NL2SQL patterns
• Connection pooling and integrated authentication
• OpenTelemetry metrics and tracing
• SDKs for Python, JavaScript/TypeScript, Go, and Java

Limitations and notes

Available tools depend on the selected database and toolset. Generic prebuilt access and custom tools use different configuration paths: prebuilt tools are selected with command-line options, while custom tools are described in YAML. Database connection settings and required credentials must be configured for the chosen source.