mcp-eval

Overview ¶

Command mcp-eval runs scripted agent tasks through the real MCP server and reports pass or fail and the estimated tokens each one costs.

By default it serves canned responses from an in-process demo API, so it needs no network, no model and no running instance. With -api and -token it drives a real control plane instead and only checks that every call succeeds. Usage: go run ./scripts/mcp-eval [-api URL -token TOKEN] [-json]