MCP Server

An MCP server is a program that exposes capabilities to AI applications through the Model Context Protocol. It offers three kinds of primitives: tools (functions the model can call, like “query this database” or “create a GitHub issue”), resources (data the host can read, like files or API responses), and prompts (reusable templates the user can invoke). Despite the name, a server isn’t necessarily a big remote machine — many are small scripts launched on your laptop by the host itself.

Servers are where MCP pays off for you as a builder. Wrap your product, database, or internal API in a server once, and every MCP host — Claude Desktop, IDEs, agent frameworks, other people’s apps — can use it immediately, with no per-host integration work. That’s why the ecosystem exploded: GitHub, AWS, Azure, Cloudflare, Stripe, and thousands of community projects ship official servers, and directories like awesome-mcp-servers catalog them. Knowing what already exists saves you from writing tools that are one npx command away.

In practice you’ll consume servers by adding an entry to your host’s config — a command like npx -y @modelcontextprotocol/server-filesystem for local servers, or a URL for remote ones — and build them with the official SDKs, where a tool is roughly a typed schema plus a handler function. A server speaks stdio when spawned locally and streamable HTTP when deployed remotely; the primitives it exposes stay identical either way.