MCP reference servers

Described in the repository as “a collection of reference implementations for the Model Context Protocol (MCP), as well as references to community-built servers and additional resources.”

The active reference servers

Seven, deliberately small and readable:

Server	What it demonstrates
Everything	A test server exercising the full protocol surface
Fetch	Retrieving web content
Filesystem	Local file access, with scoping
Git	Repository operations
Memory	Persistent storage across sessions
Sequential Thinking	Structured reasoning as a tool
Time	The minimal useful server

Thirteen further servers have been archived into a separate repository.

Why read these rather than a popular community server

The reference servers are written to be understood. If you are building your first MCP server, Filesystem and Git are the two worth reading in full — Filesystem because scoping file access correctly is the part people get wrong, and Git because it shows a server that performs real actions rather than returning data.

Licensing

Apache 2.0 for new contributions; existing code under MIT.

A note on community servers

The repository links to community-built servers. Those are not Anthropic-maintained. An MCP server runs with whatever credentials you give it and can act on your behalf — read the source of any community server before connecting it to something that matters. See the permissions entry.