vision-memory-mcp

vision-memory-mcp is a Model Context Protocol server that caches webpage and application screenshots locally to reduce token overhead and vision LLM costs for AI coding assistants. It uses perceptual hashing, CLIP embeddings, and UI state transition graphs to prevent redundant screenshot re-sending and visual hallucination loops. Built for developers using Cursor, Claude Code, Gemini, or Copilot, it operates a 4-tier caching architecture with sub-5ms performance. Features include visual spec-driven development with layout diff verification, multi-database monorepo discovery, and trajectory export for fine-tuning local models. Runs zero-infrastructure and local-first, requiring only Node.js 18.18.0+.