Connects Claude to Braintrust's evaluation and observability platform for AI applications. You get programmatic access to query your experiment results, inspection logs, datasets, and prompt templates without leaving your Claude workflow. Useful when you're debugging model behavior, analyzing evaluation runs, or pulling specific prompt versions during development. The streamable HTTP transport means you can point it at different regional API endpoints. Reach for this if you're already using Braintrust to track your AI projects and want to interrogate that data conversationally rather than switching to the web dashboard or writing SDK calls.