awesome MCP servers list
An "awesome MCP servers" list is usually a static, alphabetized GitHub README — useful for discovery, silent on quality. CheckMCP keeps the same idea of a browsable, categorized index but scores every entry 0–100 against the OWASP MCP Top 10, tool design and schema quality, and re-checks it on an ongoing basis — so its 172 catalogued servers and repositories, organized across 23 collections, are ranked, not just listed.
What a GitHub "awesome" list actually gives you
The canonical "awesome mcp servers" list — the punkpeye/awesome-mcp-servers GitHub repository — is genuinely useful infrastructure: a single README, curated by maintainers, alphabetized within categories, and currently one of the most-linked and most-scraped sources for "MCP servers" across the ecosystem.
What it doesn't give you is any signal about quality. Inclusion means someone opened a pull request that got merged; it says nothing about whether the server still works, still receives security patches, or was ever checked against a single item of the OWASP MCP Top 10. An entry can be added once and sit there for years — actively maintained or quietly archived, the README reads identically either way, because nobody's ongoing job is to re-verify hundreds of previously-merged links.
What CheckMCP measures instead
CheckMCP starts from the same instinct — a browsable, categorized index of MCP servers — but attaches a number to every entry instead of just a link. Every live endpoint is probed directly and scored 0–100 across seven pillars: security (an OWASP MCP Top 10 pass), tool design, schema and description quality, reliability, context-cost (the tokens its tool definitions cost an agent on every request), compliance and coverage.
A repo or stdio server with no public endpoint gets a parallel Repo-Quality Score instead, on maintenance, license, adoption and documentation. Hard floors apply on top of either: a hardcoded secret or a critical prompt-injection finding caps a server at grade D no matter how polished the rest of it is, and a repository its maintainer has archived caps at 54/100 regardless of its star count.
23 collections, 172 entities — organized, not just alphabetized
Rather than one flat alphabetical list, CheckMCP organizes its catalog into 23 curated collections by use case: Best MCP Servers for Databases, for Web Scraping & Crawling, for Browser Automation, for SEO, alongside a Remote MCP Servers collection covering every live hosted endpoint it tracks and an Official MCP Servers collection for first-party servers shipped by the vendor whose product they connect to.
Across those 23 collections, 172 distinct MCP servers and repositories are organized and cross-referenced — a single server can and often does appear in more than one category (a database server that's also first-party, say), which one flat alphabetical list has no way to express.
Ranked, not just listed
Inside each collection, order carries information. Most rank by CheckMCP audit score, best first, so opening "Best MCP Servers for Databases" surfaces the highest-scoring option, not whichever repository happened to be submitted earliest.
A couple invert that on purpose: the most-popular collection ranks by GitHub stars, because that's what "most popular" actually means, and the mcp-servers-to-avoid watchlist deliberately ranks worst-first and shows only entries currently carrying a failing D or F grade — the one collection on the site where landing at the top is the opposite of a recommendation.
Kept current, not a snapshot
A GitHub awesome-list updates when a human opens a pull request. CheckMCP's collections resolve against the live audit API on every page render, re-probing and re-scoring servers on an ongoing basis, so a server that stops shipping fixes, loses its license, or gets archived by its maintainer moves collections on its own.
It moves out of a "safest" list and into the avoid watchlist without anyone editing a README by hand. The catalog reflects what a server is doing right now, not what it was doing the last time someone reviewed a pull request.
The honest trade-off: audited depth vs raw breadth
None of this makes CheckMCP's directory bigger than the largest GitHub awesome-lists. punkpeye/awesome-mcp-servers alone has years of history and hundreds of contributors behind it; CheckMCP's 23 collections and 172 curated entities are real, but they're a fraction of that scale, growing as more servers get audited rather than as more pull requests get merged.
If what's needed is the single most exhaustive inventory of every MCP server that exists, a hand-curated awesome-list with a much bigger contributor base will have more of them today. If what's needed is knowing which of the servers actually under consideration are still maintained, still secure, and worth the context they cost an agent — that's a question a static list was never built to answer, and it's the one CheckMCP measures.
How CheckMCP handles it
CheckMCP organizes 172 distinct audited servers and repositories into 23 curated collections, each entry carrying its measured MCP Score rather than a position in an alphabetical list. The ranking is the audit output; a maintainer cannot buy a better place in it.