Every server is scored on security, maintenance, token efficiency, documentation, and compatibility — so you can pick one and move on.
Three servers we recommend if you're new — highest-scored, well-documented, and actively maintained.
Top 10 servers by npm weekly downloads.
Curated shortlists. Each one shows the highest-scored option for the job.
Published CVEs affecting tracked servers. Fixed means a patched version is available.
Every server is tagged with at least one category. The catalog grows daily.
Every server gets a 0–100 grade from five weighted signals. No opinion — only what we can measure from source, registry, and runtime.
CVEs, auth hygiene, sandboxing, permission model.
Commit cadence, open-issue health, release frequency.
An MCP (Model Context Protocol) server is a small program that exposes tools, data, or actions to AI assistants like Claude Desktop, Claude Code, Cursor, and Windsurf. Servers can read files, query databases, call APIs, search the web, or trigger workflows — the AI agent calls them just like a function.
Every server gets a 0–100 score across five axes: Security (CVE scanning, tool-poisoning detection, auth requirements, license), Maintenance (commit recency, GitHub stars, open issues, weekly downloads), Documentation (README quality, setup steps, examples, schema coverage), Compatibility (transports and confirmed clients), and Efficiency (total tool tokens, tokens per call). Methodology is fully public.
Start with the use-case tile that matches your work: developers usually want filesystem, GitHub, and a database server (Postgres or Supabase); productivity users want Slack, Notion, or Google Drive; AI-agent builders want web-search and memory servers. Filter by score, then check CVEs and last-commit recency before installing.
MCPpedia is free, has no paywall, and accepts community submissions. It tracks 22,413+ servers, scoring each one independently. Listings are not pay-to-play; vendors can claim and verify their servers but cannot pay for ranking.
Weekly CVE alerts, new server roundups, and MCP ecosystem insights. Free.
MCP lets your AI assistant use real tools — search Slack, manage GitHub, query databases. Set up your first server in 2 minutes.
How much of the context window tool schemas consume.
README quality, examples, transport docs, changelog.
Clients tested, transports supported, platforms.
GitHub metadata, npm/PyPI downloads, and health checks refresh on a daily cadence. CVE feeds and security advisories sync hourly. Scoring recomputes whenever the underlying signals change. Last-modified dates are exposed in the sitemap so search engines and answer engines see freshness.