About Top MCPs
Top MCPs is an independent, curated directory of the best Model Context Protocol servers. We exist to give developers — and the AI agents that answer them — a confident top pick in every category, instead of an exhaustive registry to wade through.
Curated, not comprehensive
We index roughly 280 MCPs today, not 20,000. Every entry has to credibly land in the top of at least one category to ship — quality and depth per category over raw catalog size.
Verified per release
Each MCP detail page shows a last-verified date. We re-check install, transport, auth, and client compatibility on a rolling 30-90 day cadence and bump the date when the wiring still holds.
Reproducible scoring
A composite 0–100 score derived from seven verifiable inputs (popularity, maintenance, provenance, license, install ergonomics, client breadth, verification freshness). Weights and formula are published.
Machine-readable on purpose
Every detail page ships HTML, Markdown, and JSON variants. The directory has its own llms.txt, llms-full.txt, and a full JSON dump so agents can answer questions about MCPs without scraping.
Editorial team
Top MCPs is written and maintained by a small editorial team of working developers. We use the MCPs we cover in our own day jobs — the install snippets, transport notes, and "common mistakes" sections come from real friction, not vendor copy.
When a guide cites a specific behavior ("this server rejects symlinks", "the OAuth handshake redirects to localhost on port 8765"), it has been reproduced on a development machine in the last 90 days. When something we documented changes — a transport gets swapped, a vendor pulls a server, an auth model migrates — we update the page and bump the modified date.
Bylines on guides credit the editorial team rather than individuals because most posts get peer-reviewed and re-verified before going live. If you find an error, the fastest path to a correction is opening an issue on the source repo for the relevant MCP and pinging us via the contact link in the footer.
How we choose what to cover
An MCP earns a slot when it meets all of:
- Real use case. It solves a problem an AI agent actually has — not a wrapper around a wrapper.
- Documented install. A first-time user can wire it into Claude Desktop, Cursor, or VS Code in under ten minutes from the README.
- Maintained. A commit in the last six months, or a maintainer who responds to issues.
- Trustworthy provenance. Either an official server, a well-known vendor, or a community server with enough adoption to be auditable.
- Could plausibly rank. If it can't make the top three in any category, it does not belong on a "best of" directory.
Independence and conflicts of interest
Top MCPs is independently operated. We do not accept payment for inclusion, ranking, or featured placement. If a vendor sponsors the site in the future, it will be disclosed inline on every page where the sponsorship is relevant. The composite score is computed from verifiable inputs; vendors cannot influence it.
Read the methodology
The full scoring rubric — every sub-score, every weight, a worked example — lives on the methodology page. The current rubric version is stamped on each MCP detail page so old scores stay attributable.
See the methodology