An npm report card for humans and their AI coding assistants

What it does

Packrun is a structured comparison layer over the npm registry. It groups libraries into 50+ categories—date utilities, HTTP clients, ORMs—and scores each package from 0 to 100 using a weighted formula that blends weekly downloads, gzip bundle size, commit freshness, GitHub stars, and quality markers like TypeScript and ESM support. Results are surfaced through a Next.js web app and an MCP server that lets AI assistants query the data directly.

The interesting bit

The MCP server is the real hook. Instead of treating package search as a human-only chore, Packrun exposes tools like compare_packages, find_alternatives, and validate_install directly to Cursor and Claude. The scoring weights are also published openly—freshness and quality matter more than raw popularity—which is a refreshing departure from the usual opaque “top 10” lists.

Key highlights

• Automated 0-100 scores across 50+ categories using fixed weights for downloads, bundle size, freshness, community, and quality.
• Native MCP server integration with tools for search, comparison, and install validation.
• Sub-50ms search via Typesense Cloud.
• Real-time metrics including download trends, gzip size, and days since last commit.
• Monorepo architecture with dedicated sync workers and a decisions engine for scoring logic.

Caveats

• Running your own instance requires Typesense Cloud and Redis, so it is not a lightweight self-hosted option.
• The README still contains a placeholder organization name in its clone URL, suggesting the setup docs are unfinished.
• An optional GitHub token is needed for higher rate limits, implying the sync worker can exhaust its data source quotas without one.

Verdict

Worth a look if you want data-backed package recommendations or need to give your AI agent a tool for dependency research. Skip it if you are just looking for a simple npm search replacement.