← All notes
Applied AI and ML

Connect the harness to the work

I'm in favor of teams using an existing harness, connecting their data sources, and building a skills library around the work. It takes time to get the skills right, but you can get a lot done that way.

For an internal database or a service with an API but no connector, my approach has been FastMCP Python wrappers around API calls, deployed on Modal, with Auth0 for authentication. The context in the tool definitions matters, and I don't want to build authentication myself.

Originally posted on X. Lightly edited for reading here.