Mcp Integration
July 22, 2026
Ask Your Library a Question
Markie stores your images, videos, and PDFs as plain files on your machine. Usually you find them by scrolling the gallery or the canvas, filtering by type, color, tag, or date. Markie also has a built-in MCP server, so you can ask an AI assistant to search your library in natural language instead.
What MCP Does Here
MCP (Model Context Protocol) is a way for AI assistants like Claude to connect to tools and data. Markie's MCP server exposes your library so an assistant can search and retrieve your assets on request. You describe what you're looking for; the assistant looks through your saved files and returns the matches.
No separate app to open. No filter menu to configure. You ask, in your own words, and the assistant reads back what's in your library.
Why This Matters for a Personal Archive
A saved-asset library grows fast. Screenshots, reference photos, PDFs, video clips: after a few months of one-click saving from the Chrome extension, a plain grid or a folder tree stops being enough. Manual tags and the gallery filters still work well for browsing. The MCP server adds a different way in: instead of remembering which tag you used or which month you saved something, you describe it and let the assistant search.
This sits alongside Markie's optional AI tagging and semantic search, both of which run locally via Ollama if you turn them on. The MCP server doesn't replace them. It gives an outside assistant a route into the same library, using whatever tags, descriptions, and metadata already exist there.
How It Connects
The MCP server runs via npx -y @vitgranen/markie-mcp, published on npm. Point a compatible AI assistant at it, and it can query your library the same way it would query any other connected tool.
It's included at no extra cost on every tier: Free and Pro both get the full MCP server, with no separate fee. The only limit on the Free tier is the 500-save cap; Pro removes that.
Your files still never leave your machine. The MCP server is a local connection between your library and an assistant you choose to connect it to. Markie has no servers and no accounts, and this doesn't change that.
For Developers, Too
If you're building your own tools on top of a Markie library, the MCP server sits next to the developer SDK (@vitgranen/markie-sdk), also included at no extra cost. Between the two, you can wire an AI assistant into your library or build a custom frontend against it.
Try It
Set up the MCP server and start asking your library questions directly.