CatoBot AutoExperiment is a versatile Model Context Protocol (MCP) server designed to enable AI agents to perform autonomous experimentation across any domain. It generalizes the 'modify something → run it → measure a result → keep or discard → repeat' pattern, exposing it as a standard set of MCP tools. Users define their experiment's domain, including what gets modified, how it runs, and what gets measured, entirely through a JSON configuration file. The server tracks progress, manages file changes, executes commands, extracts metrics, and handles version control via Git, ensuring reproducible results and enabling intelligent agents to iteratively optimize performance for a single, defined metric.
주요 기능
01Domain-agnostic configuration via JSON for flexible experimentation setups
02Metric-driven keep/rollback decisions based on configurable policies
03Reproducible experiment history maintained through Git commits
04Standardized MCP tools for seamless AI agent interaction and control
05Support for shell, external, and hybrid experiment execution modes
061 GitHub stars
사용 사례
01Automating iterative optimization tasks driven by AI agents
02Experimenting with and improving performance of training scripts, benchmarks, or simulations
03Generalizing 'autoresearch' patterns for diverse domains
