MCP Directory
Intent search over awesome-mcp-servers, with embeddings in pgvector.
~2,800 MCP servers indexed with embeddings (pgvector); weekly re-sync
Production artifacts running in this repo — each with a concrete number, not a claim.
Intent search over awesome-mcp-servers, with embeddings in pgvector.
~2,800 MCP servers indexed with embeddings (pgvector); weekly re-sync
The same search pattern over the skills.sh corpus.
~400 skills indexed (ingest cap --limit 400)
Hybrid RAG: facet filter + semantic ranking.
~14,000 templates in the consultant corpus (2,800 curated + 10,034 marketplace + 1,240 software); 3-tier setup grid in code
Scheduled ingest and freshness alerts for the MCP and skills catalogs.
4 GitHub Actions (MCP resync, skills resync, daily watchdog, email alert)
Scrape → dedup → curate → rewrite → generate via OpenRouter, with cost controls.
10 sources, max 4 items/day; default model is OpenRouter free-tier (USD 0/month at current model)
Public API proxies keep secrets off the client, with in-process rate limits and a CORS allowlist.
30 req/min on /api/mcp-search; 3 req/min on /api/recommend; 4 production origins on the CORS allowlist
Index repo for a personal series — not created yet.
TODO — repo does not exist yet