datasette-mcp 0.2

Simon Willison's datasette-mcp 0.2 changes how database results are presented to AI tools. The plugin adds an MCP server, a standard way to expose tools to AI agents, to a Datasette database instance. In this release, the rows returned by its execute_sql tool are objects rather than arrays of values, which associates each value with a named field instead of requiring the model to remember which position matches which column. Willison says this should help weaker models avoid losing track of that mapping. Release note.

In an earlier post on stateless MCP, Willison described the plugin's tools as listing databases, inspecting a database's schema and running SQL queries, with queries read-only at that time. The 0.2 update changes the structure of query results, not that capability, and the release does not quantify any accuracy gain. It is also the plugin's first non-alpha release. Original plugin explanation.