About this directory
An independent directory of MCP servers, with a written review behind every entry. No sponsorships, no paid placement, and no server listed without a reason you can read.
Last updated · 5 September 2026
What this is
This site catalogues MCP servers — the Model Context Protocol processes that give an AI agent access to a database, an API, a filesystem or a SaaS account. A server decides what an agent can touch, which makes “is this one safe to connect?” a more consequential question than a star count can answer.
So each entry carries a profile: what the server's tools actually do, what permissions it needs, how to configure it in your client, and a scored review with the reasoning attached. Every profile exists in English and Chinese.
How an entry gets here
Candidate repositories are discovered on GitHub by rule, not by submission queue or payment. Each is then read against its own README, its tool declarations, its commit history and outside sources, and written up as a profile.
The write-up is produced by analysis assisted by automated tooling and reviewed before publication. Where the evidence is thin, the profile says so rather than filling the gap with confident-sounding prose.
The score
Every profile carries an FMRS score built from five dimensions, each shown on the page with its own reasoning. It is not a popularity number.
- Reliability — does it start consistently, complete the MCP handshake, and behave as its tool declarations claim?
- Security and permissions — are permissions, credentials, data boundaries and side effects clear, and do they follow least privilege?
- Maintenance — are releases, commits, issue responses and dependency updates still active?
- Documentation — are setup, auth, tool parameters, limits and troubleshooting documented with verifiable detail?
- Setup experience — how many steps, and how complex, from dependencies to a working connection in a major client?
Independence
This directory is not affiliated with, endorsed by, or operated by Anthropic or any vendor whose servers appear here. Model Context Protocol and MCP are used descriptively to name the protocol.
Nothing on this site is paid for. There is no advertising, no sponsored ranking, and no way to buy a listing or a higher score. No publisher sees a profile before it goes live.
Corrections and removals
If you maintain a listed repository and a profile is wrong, out of date, or you would rather not be listed at all, write to [email protected] with the URL. Corrections are made on the merits, and removal requests from a repository's maintainer are honoured without argument.
Sister directories
The same team runs FollowSkills for Claude Agent Skills and FollowAgents for AI agents, on the same review principles.