marc-shade/world-intel-mcp

Implements a distributed intelligence architecture with **Qdrant vector store + FastEmbed** for semantic search across 110+ accumulated data sources, enabling natural-language queries like *"military activity near Taiwan"* without indexed databases. Uses **async Fetcher with per-source circuit breakers, SQLite WAL cache with TTL, and background worker queue** for non-blocking data ingestion — designed for AI agents via stdio MCP protocol, with optional live Leaflet ops-center dashboard (SSE-streamed 35+ feeds) and PDF/HTML report generation. All data sourced from free public APIs (Yahoo Finance, SEC EDGAR, GDELT, OpenSky, USGS, etc.) with no paid subscriptions; 43 parallel data collectors populate the vector store on 5-minute intervals.