"author:gbesse" matching MCP servers:

Enables clients to submit an execution-trace case and receive a structured, read-only conditional compensation plan that models ambiguous outcomes, requires observations, and replays each branch without executing external actions.

Runs a local stdio MCP server that lets you evaluate an AI agent's business outcome against four synthetic scenarios, including a duplicate-free quote follow-up, an unsent invoice reminder, client-file separation, and a stockout. It scores the run from the final state and recorded attempts rather than the agent's text, returning a pass or fail exit code.