MediaLibrary

A personal media tracking system that aggregates books, movies, TV shows, podcasts, music, and recipes into a single searchable database with AI-accessible context.

Overview

MediaLibrary was built to solve a specific problem: AI assistants have no persistent memory of what you’re reading, watching, or listening to. By tracking media consumption in a structured database exposed through MCP, every PAI session knows what books I’m currently reading and what music I listen to.

Key Features

• Multi-source ingestion - Spotify listening history (automated via launchd), Libby/OverDrive CSV exports for library books, manual entries via CLI
• Unified database - SQLite-backed storage for 375+ media items across all types
• Personal API integration - MCP endpoints for current media, taste profiles, and recommendations
• CLI management - Add, update, search, and browse media from the command line
• Automated sync - Launchd-scheduled Spotify history imports every 6 hours

Architecture

Spotify API ──┐
Libby CSV ────┤──> CLI Ingest ──> SQLite DB ──> Personal API (MCP)
Manual entry ─┘                                  └──> AI Context

Integrations

Spotify Sync

Automated via launchd scheduler, pulls recent listening history and stores tracks with metadata (artist, album, play count, timestamps).

Libby/OverDrive Import

Parses CSV exports from the Libby library app, importing borrowed and completed books with metadata.

What I Learned

• Spotify API rate limits - Need careful pagination and caching for listening history
• Data normalization - Merging the same book from multiple sources (manual entry vs. Libby import) requires fuzzy matching
• MCP as a personal data layer - Exposing personal data through MCP makes every AI session contextually aware