mcp-eval

Run the same code-research task through two isolated sub-agents — one with the code-index MCP, one with only Read/Grep/Glob — and produce a side-by-side comparison of token usage, tool calls, latency, and answer quality. Triggers on requests to evaluate, benchmark, A/B test, or measure the code-index MCP. Use when the user says 'evaluate the MCP on [task]', 'is the index worth it', 'how much does the MCP save',…