What a database MCP connector should control.
A database MCP server turns database operations into named tools an AI client can call. A useful production connector also needs identity, workspace permissions, destination checks, credential isolation, and strict query limits.
Data Connector's tool boundary
list_data_sourcesreturns live database sources visible to the authorized workspace.search_data_catalog,describe_data_source, andquery_data_sourceprovide bounded external database access.list_tables,describe_table, andexecute_sql_queryprovide bounded access to Loam-managed data.
The hosted MCP endpoint is https://dataconnector.ai/api/mcp. It uses OAuth, so Claude receives scoped access rather than a raw database password.
Supported sources
Live external connections currently support PostgreSQL. SQLite, CSV, and Parquet files can be imported into Loam and queried as Loam-managed tables. MySQL, SQL Server, SaaS apps, and arbitrary APIs are not available yet.
Database MCP versus a direct connection
A direct connection gives the client a credential and broad protocol access. Data Connector keeps live credentials server-side and exposes a fixed read-only tool surface with workspace checks on every request.
Frequently asked questions
What is a database MCP server?
It exposes named database discovery and query tools that an MCP client such as Claude can call after authorization.
Which tools does Data Connector expose?
It exposes four tools for live sources and three tools for Loam-managed tables. All seven are read-only.
Does Data Connector support MySQL?
Not yet. Live external connections currently support PostgreSQL, while SQLite uses Loam's file import path.