Nornic
CLI · MCP · API
06Chapter · CLI · MCP · API
3 min read
cli-mcp-apimcpecosystemserverstools

The MCP Ecosystem: Real Servers and Clients

The last lesson explained what MCP is. This one answers the question that comes right after: what is actually out there? The quiet truth is you rarely build a server — thousands already exist, and connecting one is mostly config.

Last updated ·
Share

You almost never write an MCP server. You pick one off the shelf — so it's worth knowing what's on the shelf, and how to read it.

By the end of this lesson

You'll know the two roles that fill the MCP ecosystem, the official servers worth starting with, where to find hundreds more, and why connecting one is configuration rather than a coding project.

Because MCP is a protocol, it has an ecosystem of things that speak it — and the whole ecosystem is just two roles. Servers wrap a capability — your files, a database, GitHub — and expose it. Clients (also called hosts) are the apps a model runs inside that connect to those servers. The job is simply matching a client you already use to the servers you need.

The official reference servers

The protocol's own team maintains a small set of servers as worked examples — clean, well-built, and the right place to start. A few you'll meet immediately:

official reference servers
filesystemread & write files in a folder you allow
gitread a repository — history, diffs, branches
fetchpull a web page in and hand it to the model
memorya persistent store the model can write notes to

Each one is small and does a single, legible thing — the same Unix instinct from earlier in the course, applied to AI tools. Reading their descriptions is itself a good way to build intuition for what a tool should be.

And the hundreds more

Past the official set sits a much larger community catalog. The well-known index is the awesome-mcp-servers list, with servers for Slack, Postgres, Google Drive, Stripe, and hundreds of others. Browsing it is the fastest way I know to see what “give the model a tool” looks like in the wild — and to discover that someone has probably already built the server you were about to write.

Connecting a tool is configuration, not a project.

On the other side of the catalog, any MCP-capable client connects to all of them — Claude Desktop, the Claude and OpenAI developer tools, and a growing set of editors. That's the promise from the protocol lesson made concrete: write the server once, and every client can speak to it. You don't build integrations anymore. You pick them.

filesystem GitHub database web · HTTP MCP client one protocol
One client speaks one protocol to many servers. Want a new capability? Plug in a server instead of writing fresh glue every time.

Try this

Open the awesome-mcp-servers list and find a server for a tool you actually use. Read its description and the tools it exposes. That five-minute browse will tell you more about what's possible with MCP than any amount of theory.

Sources — Model Context Protocol, example servers and modelcontextprotocol/servers; awesome-mcp-servers.

0

Tap to appreciate

Was this chapter helpful?

Comments

New chapters land here as I learn them. Want the next one?