Overview
PAI Intel MCP is a personal cybersecurity intelligence pipeline built as a Model Context Protocol (MCP) server. It reads RSS and Atom feeds from IAM, AI security, and Agent/MCP sources, scores each item across 6 dimensions relevant to my work, and surfaces high-signal content into a Notion review queue and Astro draft files.
Nothing publishes without my explicit approval. The pipeline finds the signal — I decide what’s worth sharing.
Architecture
RSS/Atom feeds → Ingest → Deduplicate → Score (LLM) → Brief → Notion queue → Astro drafts
↓
NOESIS (MCP tools)MCP Server — exposes 5 resources (intel://latest, intel://top-signals, intel://weekly-brief, intel://source-health, intel://drafts/astro) and 8 tools callable directly from Claude Code.
Scoring — each item is scored by an LLM across 6 dimensions (0–5 each): IAM relevance, AI security relevance, Agent/MCP relevance, Actionability, Content potential, Strategic relevance. Items scoring ≥ 20/30 aggregate surface as top signals.
Brief generation — daily briefs include Top Signals, LinkedIn post ideas, and Teaching Lab ideas. Weekly briefs group by domain with a strategic pick recommendation.
Why I Built This
I read a lot of security content. Most of it is noise. I needed a system that could filter the 300+ items/week coming through my RSS feeds down to the 5–10 that actually matter for my work in IAM and AI security — and that could generate draft content from those signals without me starting from a blank page.
The MCP interface means I can ask NOESIS “what are the top IAM signals this week?” and get an actionable answer directly in my editor.
Status
Pipeline is operational. Notion integration and Astro draft generation tested. Running on-demand via bun run pipeline; cron scheduling next.