[MCP Server] skills-but-for-dev: Verify GitHub star velocity, PyPI adoption & VRAM before Cursor writes code

Showcase Built for Cursor
1

Hey Cursor community! :waving_hand:

Whenever I ask Cursor to build a complex feature (like audio diarization, PDF parsing, or local RAG), it often defaults to libraries from its static training cutoff—some of which haven’t had a commit in 18+ months.

To solve this, I built an open-source FastMCP server called skills-but-for-dev.

What it does:

Before Cursor writes code, it queries live metrics across GitHub, PyPI, and Hugging Face:

  • Dual-Bucket Search: Balances proven industry leaders (stars:>500) with fast-rising breakout repos (stars:75..3500, pushed < 30 days).
  • Real-World PyPI Adoption: Checks 30-day downloads via pypistats.org to guard against GitHub star inflation.
  • Star Velocity & Acceleration: Computes 4-week moving average ($v_4$) and acceleration ratio ($v_4 / v_{12}$).
  • License & VRAM Check: Flags GPL/non-commercial licenses and estimates minimum GPU VRAM for Hugging Face / GGUF models.

Quick Setup in .cursor/mcp.json:

{
  "mcpServers": {
    "skills-but-for-dev": {
      "command": "uvx",
      "args": ["skills-but-for-dev"]
    }
  }
}

(Auto-detects your gh auth token if you use GitHub CLI, so zero API key config needed!)

Would love your feedback!