How to manage local image libraries with Cursor and Codex
Stop uploading images manually. Learn how to use MCP tools to let your AI agent automatically tag, sort, and retrieve visual assets directly from your disk.
The problem with manual image management
When working with AI agents like Cursor or Claude Desktop, dealing with visual assets usually involves a lot of friction. You have to manually drag and drop images into the chat, keep track of what you've uploaded, and there's no programmatic way for the agent to browse your existing assets.
Enter Magpie and MCP
With the Model Context Protocol (MCP), we can give AI agents direct, structured access to our local file system. Magpie acts as a visual memory layer, providing tools that let your agent search, tag, and organize images autonomously.
// Example: Asking Cursor to organize a folder
"Please look at all images in ~/Downloads/raw_assets,
tag them based on their content, and find any duplicates."Because Magpie runs locally, your images never leave your machine unless you explicitly share them with the LLM API during a specific task.