evalpair for Claude Code

io.github.srikarmk/evalpair

Run LLM eval suites on local or hosted models; compare runs with intervals and a noise floor.

client:Claude Code transport:stdio runtime:pypi

Install evalpair in Claude Code

run in your project directory

claude mcp add evalpair -e EVALPAIR_HOME=<EVALPAIR_HOME> -e EVALPAIR_API_KEY=<EVALPAIR_API_KEY> -- uvx evalpair

Adds it for this project only. Append --scope user to make it available everywhere. Claude Code docs