datasette-mcp 0.2
Simon Willison’s Weblog Simon Willison
datasette-mcp 0.2 changes SQL results to rows of objects, not arrays. That should make weaker models less likely to mix up columns, and it's the plugin's first non-alpha release.
Based on reporting by Simon Willison’s Weblog, Simon Willison — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Simon Willison has shipped datasette-mcp 0.2, and the big change is simple: the rows returned by execute_sql now come back as an array of objects instead of an array of arrays.
That sounds small until you remember what it’s for. With positional arrays, a model has to keep track of which slot means what. Miss that thread once and the whole result gets mushy. Object-shaped rows give the model named fields to hold onto, which should make the output easier to read and less fragile for weaker models.
The release also bumps the plugin’s requirement to mcp>=2.1.1. And this one crosses a line: it’s the first non-alpha release of datasette-mcp.
Willison says he’s comfortable calling it ready because he’s been using it quite a bit himself. That’s the kind of endorsement that matters more than a polished changelog. A lot of AI tooling still ships with the vibe of “please don’t look too closely.” This one is moving toward something sturdier.
My take — AI-written commentary, not fact-checked reporting
This is the right kind of AI product work: less flashy, more legible. Object-shaped rows are boring in the best way, because boring is what you want when a model is trying not to lose the plot. The industry could use fewer grand demos and more changes that stop small mistakes from becoming nonsense.
Read more about this at: Simon Willison’s Weblog
