Credit Optimizer v5 for Manus AI

This server tackles Manus AI's credit burn problem by intercepting prompts and routing them through cheaper execution paths. It classifies intent across 12 categories (code, research, creative, data, translation) and downgrades from Max to Standard mode when complexity doesn't justify the premium, routes simple Q&A to free Chat Mode instead of paid Agent Mode, and applies context compression to prevent exponential token growth. The MCP version exposes the routing logic as callable tools. The paid Manus Skill version runs automatically on every prompt without manual invocation. The author claims 47% average savings across 53 audited scenarios with zero quality loss, enforced by a hardcoded veto rule that skips optimizations if output degrades. Pays for itself in roughly 36 prompts if the math holds.