The README is direct about what this is: a server that hands database access to a model. It tells you to create a dedicated OceanBase user with minimal permissions, never to use root or an administrative account, to restrict access to the operations actually needed, and to consider query whitelisting before production. Take that literally — the connection you configure defines the blast radius, and nothing in the server narrows it for you.
An MCP server for OceanBase that exposes tables as resources and lets a model read data and execute SQL against them, with connection details supplied through environment variables rather than embedded in prompts.
- Available OceanBase tables surfaced as resources, so the schema is discoverable without a query
- Table contents readable directly
- SQL execution with proper error handling, so a malformed query returns a message rather than a stack trace
- Comprehensive logging of database operations, which is what makes an audit trail possible
An OceanBase database and a user for it. Five environment variables: `OB_HOST`, `OB_USER`, `OB_PASSWORD`, `OB_DATABASE`, and `OB_PORT` which defaults to 2881 if unset. The documented install line names `oceanbase-mcp-server`, which PyPI does not have; the project is `oceanbase_mcp_server` (0.1.0 in pyproject) and runs as `python -m oceanbase_mcp_server` or through uv against a checkout. Apache licensed.