Consumer Agent Testing Tool (CATT)

About

CATT runs scripted or model-driven surrogate assistants (Muse or Instinct emulation) against an MCP server, MCP app, website or HTTP API. It checks what the service actually did, records tool calls and responses as evidence, and compares results across builds, locally or in CI. It labels each assumed assistant behavior as documented, unofficial, assumed or unknown, and a passing test does not certify compatibility with a hosted assistant.