MCP servers
Find servers by what they need to do, not by repository language. Every entry points back to its original source.
How we label these entries
Each card shows where the entry came from, the repository path where we can establish one, when we last saw it, and the public star count GitHub reports. A field marked not held means our database holds no value for that entry - it is not a statement about the project and not a verdict. We do not label repositories with pricing, interface language or registration status: these are code repositories, not product sites, so we do not measure those facts and will not guess them. This directory only lists entries that carry a star count (5,726 of 10,162 site-wide entries, 56.3%); where there is no star count, nothing is shown - we do not print a zero or a placeholder. Links are checked on a rolling basis · last sweep 2026-11-14.
Filtering happens in your browser as you type. No page reload, the URL does not change.
Showing 1–48 of 2,857 entries listed here (page 1 of 60)
Common questions about MCP servers
What is an MCP server?
A program that implements the Model Context Protocol, exposing a capability — files, a database, web search — as a tool an AI assistant can call.
How do I install one?
Most entries are open-source projects you install from their README. Our entry pages carry the original source link, whether a README exists, and when we last saw it. An install command is shown only when it could be confirmed; when it could not, the page says so rather than printing a command that will fail.
Do they cost money?
The entries are public projects. Third-party services behind them (cloud APIs, paid data sources) may need your own API key or subscription — that is the provider charging, not us.
Local or remote?
A local server runs on your machine over stdio: you control the data, but you install and run it. A remote one is hosted by someone else over HTTP: simpler to configure, but it depends on their uptime and reachability.
Which clients work with them?
Claude Desktop, Claude Code, Cursor, VS Code, Codex and any other MCP-capable client. The configuration format differs per client; the project README has the exact shape.
How do I judge whether one is safe?
Before installing: is the source public, when was the last commit, how many dependencies, what permissions does it ask for, where does data go. For anything touching the filesystem, a database or a shell, grant the minimum and rotate credentials. We do not certify servers and do not show a "verified" badge — that would need checks we do not perform.
How do you decide remote vs local?
Three kinds of checkable evidence, strongest first: (1) the official MCP Registry records a remote endpoint (streamable-http / sse) for this repository; (2) the entry URL is itself an MCP endpoint (/mcp, /sse); (3) the install command hands you an endpoint URL. Local is one thing only: an install command that runs on your own machine (npx / uvx / pip / docker run). Everything else is unknown rather than guessed - a name containing "remote" or a README mentioning streamable is not evidence.
Is "remote servers" a 14th top category?
No. The 13 top categories are fields; remote is a parallel filter, and one entry can be both. Counting it as a 14th category would double-count entries and make the 13 category totals disagree with the catalogue total, so it stands on its own: its number equals the cards on its page and is not part of the category tally.
What we actually check
That the source link resolves, when we last saw it, whether a README exists, and the public star count. All four are checkable on the page itself. We do not label client support or self-hosting — the database has no such field, and a badge for data we do not have would be invented. The one-line summary on each card is our plain-English rewrite of the project’s own description; the original wording is one click away in the source link.
Found something missing or wrong?
Tell us and we re-check it against the source before it goes in.
Contact us ↗