AI Stack Observatory

Know what exists, know what you already have, understand the overlap, and know what is worth your attention.

This is not another AI tools directory. Paste a GitHub URL. In under 60 seconds you get a USE / EVALUATE / WATCH / IGNORE / REPLACE / RETIRE recommendation against your stack — with overlap %, closest existing tool, and “reevaluate if” triggers.

The north-star reply is: Interesting, but you don't need this.

Screenshots

Demo

pnpm install
pnpm test
pnpm dev

Open http://localhost:3000 then Open demo stack. Try Analyze with https://github.com/example/viking — you should see WATCH, 74% overlap with Graft, 87% confidence.

Contributor Postgres (not used on the operator Mac):

docker compose up -d

CLI

pnpm observatory inspect https://github.com/example/viking

MCP

Real Model Context Protocol server (Streamable HTTP + stdio), not a homemade REST wrapper.

Remote (K3s A): http://192.168.68.57:31501/observatory/mcp with Authorization: Bearer aso_demo.

{
  "mcpServers": {
    "ai-stack-observatory": {
      "url": "http://192.168.68.57:31501/observatory/mcp",
      "headers": { "Authorization": "Bearer aso_demo" }
    }
  }
}

Local stdio from this repo: pnpm observatory mcp.

Tools: search_technology, get_stack, compare_tools, analyze_repository, find_alternatives, get_recommendation, get_subscription_cost, get_watchlist.

Data

Catalog seed reuses the AI Agent Infrastructure Landscape (CC-BY-4.0) and the official MCP Registry. See data/ATTRIBUTION.md.

Architecture

Next.js App Router + Drizzle/Postgres + recommendation engine in packages/core. Deploy: OpenNext on Cloudflare. UAT: K3s A/B (observatory-a / observatory-b).

License

MIT. Landscape data remains CC-BY-4.0.