Great write-up, but looking at the results, it seems the massive 78% token reduction comes primarily from the execute tool batching actions, rather than DevTools being inherently lighter than Playwright. If someone added a similar sandboxed execute tool to Playwright-MCP to allow batching, wouldn't the token usage and turn count be nearly identical? I'm trying to understand if the protocol (CDP) is the real winner here, or just the smart API design of your MCP