MCP Official Servers
The canonical reference MCP servers — filesystem, fetch, memory, SQLite, and more, straight from the spec maintainers.
About
modelcontextprotocol/servers (91k stars) is the official collection of reference MCP server implementations maintained alongside the Model Context Protocol spec: filesystem, fetch, memory, sequential-thinking, SQLite, and others.
These are the safe defaults the ecosystem standardizes on — intentionally minimal in places, which is a feature for security review, not a gap. Licenses are mixed per-server (predominantly MIT), so check the individual server you deploy. The repo also links all ten official MCP SDKs (TypeScript, Python, Go, Rust, Java, Kotlin, C#, PHP, Ruby, Swift).
What it can do
- Reference implementations for the most common server patterns (files, fetch, memory, git, sqlite)
- The canonical examples the whole MCP ecosystem standardizes against
- Links to all ten official MCP SDKs for building your own servers/clients
- Minimal, auditable codebases — good starting points for security review
- Active maintenance from the spec stewards (pushed Oct 2026)
Setup
- Pick a server from the repo (e.g. filesystem, fetch, memory, sqlite).
- Run it via npx/uvx exactly as its README shows — no install step needed for most.
- Add the command to your MCP client's config (Claude Code, Cursor, etc.).
- Scope filesystem servers to specific directories using their path arguments.
- For production use, prefer the hosted/packaged variants over npx where available.
Prerequisites
Node.js (npx) or uv (uvx) depending on the server; an MCP-compatible client.
Use cases
Works with
Trust score — 9/10
Scored 9/10 (verified tier). At 91k stars this is the most-starred MCP repo in existence and the canonical reference from the spec maintainers — the safest default in the category. Docked one point: licenses are mixed per-server (mostly MIT, but verify each), and reference servers are deliberately minimal, so production hardening (auth, sandboxing, resource limits) is on you.
★ 91,183 GitHub stars
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 9.2/10
- Maintenance · recency, releases, issue load
- 9.2/10
- Openness · license + docs
- 8.6/10
- Manual review · sanity read + red-flag check
- 9.0/10
- · 91,183 GitHub stars (log-scaled vs catalog peers)
- · last push 1d ago; 9 releases/12mo; 488 open issues
- · manual review: canonical reference servers (91k stars), the safe default; minimal by design; production hardening is on you
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.