One Schema Change in datasette-mcp 0.2 Fixes a Model's Column Confusion
datasette-mcp 0.2, the first non-alpha release of Simon Willison's Datasette MCP plugin, changes execute_sql to return rows as an array of objects instead of an array of arrays.
What it is
datasette-mcp is Simon Willison's Model Context Protocol server plugin for Datasette, now at its first non-alpha release.
What it does
Version 0.2 changes the `execute_sql` tool's `rows` field from an array of arrays to an array of objects, so each row carries its column names instead of relying on positional order, and bumps its dependency to `mcp>=2.1.1`.
Why it matters
Positional arrays force a model to track which index maps to which column across a whole conversation; naming the columns removes that failure mode entirely, which matters a lot for weaker or cheaper models running SQL through an MCP tool.
How to use it
Install or upgrade the `datasette-mcp` plugin to 0.2 via pip or the Datasette plugin installer; any client parsing `execute_sql` output as positional arrays needs updating for the new object shape.