Recent content by ellievis

With experiments, I'd keep a basic version that already works and compare every change against it. That gives you a proper baseline instead of relying on memory. If it works twice in a row, then I would start expanding the test.

This tutorial shows a simple repeatable way to work with MCP while focusing on testing. Step 1: Define the result you want. Write down the output format, quality requirements, and any limits before you start. Step 2: Prepare a small representative example. Test with something easy to inspect...