evalgate MCP
v0.4.1io.github.ipezygj/evalgate
Statistical checks an agent runs before trusting an AI eval number (is #1 real, judge bias, more).
transport:stdio
runtime:pypi
Target client
run in your project directory
claude mcp add evalgate -- uvx eval-integrity
Adds it for this project only. Append --scope user to make it available everywhere. Claude Code docs
This listing does not declare its tools. Connect the server and your client will discover them on the handshake.