Windows Preview available. Pro checkout is not open yet. Plans and capability details →

MCP, with a visible boundary

Your assistant asks.
Your local engine works.

Connect an MCP-compatible client to the RowCove stdio server. Keep data access explicit, results bounded and every step reviewable.

A useful request
Inspect this dataset's schema.
Choose only the required columns.
Query one date partition first.
Show source provenance for findings.
Do not export or edit anything.

Tools for the actual data workflow.

Discover shards, inspect schemas, read selected database tables, query data and inspect source records. Review tool definitions and JSON schemas in the MCP settings before a test call.

Long-running jobs expose progress and cancellation. Mutating operations retain their confirmation and authorization requirements.

Setup in four steps

  1. Activate Pro. In Account and plan, use your purchase email and the one-use email code.
  2. Open MCP settings. Enable access, review authorized roots and copy the generated client configuration.
  3. Configure your client. Paste the generated configuration into its MCP server settings. Use the executable and paths supplied by your installation.
  4. Test a read. Start with license_status, then inspect a small permitted source. Review the tool result, wire log and history.

Understand the privacy boundary

The local server reads authorized local sources. The tool response goes to the AI client you connect. If that client uses a cloud model, returned rows can reach its provider.

Authorize only necessary directories. Avoid broad drive roots, request projected columns and bounded pages, and review results before sharing. Rotating the MCP token invalidates older client authorization.

Read the detailed MCP tutorial
Pro licensing and MCP authorization are separate checks. A valid Pro license does not grant unrestricted filesystem access, and an MCP token does not grant Pro.