vision-memory-mcp
Active·★ 48·NOASSERTION·Updated 2026-10-08
vision-memory-mcp is a local-first Model Context Protocol (MCP) server and CLI tool designed to enhance AI coding assistants. It achieves this by providing visual state caching, using techniques like perceptual hashing and local CLIP embeddings, to significantly reduce repetitive vision LLM calls and improve efficiency.
#Visual Caching#AI Assistant#Perceptual Hashing#CLIP Embeddings#Video Analysis#Element Grounding#UI Automation#Local-First
01
Features
01Provides sub-5ms perceptual visual caching using dHash and local CLIP, significantly reducing vision LLM API calls.
02Ingests WebM/MP4 videos to create searchable keyframe visual states and transition graphs.
03Maps screen elements to CSS selectors for action prediction and supports visual spec contracts for regression testing.
04Ensures complete local processing and storage of all visual data, including CLIP models, with no cloud telemetry.
02
Compatibility
Node.js
Runtime
Verified via docs
TypeScript
Language
Verified via docs
03
Quick start
1
$ npm install -g @putervision/vision-memory-mcp
04
Use cases
↳Providing visual context and memory to AI coding assistants like Cursor or Copilot to reduce redundant vision LLM calls.
↳Ingesting E2E test recordings to analyze UI state transitions, detect visual regressions, and verify design contracts.
↳Enabling deterministic UI interaction and action prediction for automated agents, and generating datasets for multimodal fine-tuning.
↳Managing visual states and context locally for applications requiring high privacy or offline capabilities.
05
Alternatives
OpenChatCut★ 2.6k
Local-first conversational AI video editor with a professional multi-track timeline, Agent Skills, MCP integration, and Remotion-powered rendering.
ComfyUI-Copilot★ 5.6k
An AI-powered custom node for ComfyUI designed to enhance workflow automation and provide intelligent assistance
Peekaboo★ 5.4k
Peekaboo is a macOS CLI & optional MCP server that enables AI agents to capture screenshots of applications, or the entire system, with optional visual question answering through local or remote AI models.
Concat★ 10.5k
The truly free, and open-source cross-platform CapCut replacement (supports MCPs).
blenderwright★ 175
MCP server for Blender. 191 tools for modelling, materials, rendering and more, driven by Claude, Cursor, Codex, Ollama, llama.cpp, or any MCP client.
Comments
Log in to leave a comment
No comments yet. Be the first!