SQL Preview

You'd reach for this when you need Claude to query databases without giving it free reign to run arbitrary SQL. It exposes governed query execution and schema inspection capabilities, letting you define guardrails around what database operations an agent can perform. The governance layer is the key piece here, useful when you're building agents that need to pull data or explore table structures but shouldn't have direct SQL access. Works over stdio transport, so it runs as a local process that your MCP client connects to. Think of it as a controlled database interface for AI agents rather than handing over raw connection strings.