Show HN: MFKVault — find AI skills faster than GitHub search

1. Why GitHub search is noisy

Searching "MCP server" or "Claude skill" on GitHub returns thousands of repos: half tutorials, half empty starter templates, scattered licenses, no install commands, and no way to tell what actually runs. Discoverability is the blocker — not the lack of code.

2. How MFKVault helps

• Search across GitHub, MCP directories, npm, PyPI, and curated awesome lists in one place.
• Every result shows source URL, author, license, safety status, and last-checked date.
• One-click install command copy. No auto-run.
• Save skills, mark them works/broken, add private notes.
• Opt-in weekly digest of new matching skills.

3. What works today

• Crawlers ingest only OSS-licensed sources and tag everything with SPDX.
• Dev.to/blog/news posts are imported as pending_review — never installable.
• Saved skills dashboard with works/broken/notes tracking.
• Search filters: source, safety, ready-to-install, license known, recently updated.
• Weekly alert digest pipeline with idempotent send log (no duplicate emails).

4. What is still rough

• License coverage varies — many older repos have Unknown. We show that explicitly instead of guessing.
• Install commands aren't normalized across ecosystems yet. Some entries show "Install instructions unavailable — review source".
• Quality scores are heuristic and require manual review for the top tier.
• HuggingFace spaces and proprietary repos are not installable by default.

5. Try these searches