I built a faster drop-in replacement for Playwright MCP — 41% faster, 6 less context, cookie import

I built pilot because I was tired of watching my AI agent wait. Hundreds of browser calls per session, each one taking 100-200ms through existing MCP browser servers. That adds up to 20+ seconds of pure latency for a typical automation flow. After running benchmarks across 5 real tasks with Claude Code as the runtime: pilot @playwright/mcp Wall time 25s 43s — 41% faster Tool result size 5,230 chars 9,165 chars — 43% smaller Cost/task $0.107 $0.124 — 13% cheaper Success rate 5/5 4/5 The Architecture Insight Most MCP browser servers run Playwright as a separate process, talking over HTTP or WebSocket. Every action pays for serialization, network round-trip, and deserialization. Pilot takes a simpler approach: run Playwright in the same Node.js process as the MCP server. No HTTP layer. ~5ms per action vs ~100-200ms. The Context Size Problem Playwright MCP dumps a full snapshot on every navigate() — typically 58K chars. Pilot returns ~1K chars (top 20 interactive elements by default). On a