مضيف واحد وخوادم متعددة

So far you have run a single server. But a real host like Claude Desktop can connect to many servers at once, each adding its own set of skills. 🔌

You could cram everything into one server, but small focused servers are easier to build, test, and reuse. One does files, another does GitHub, another does search.

The host is the conductor. It launches each server, opens a client for it, and gathers all their tools into one menu the model can choose from.

Remember the one-link rule: the host spins up a separate client for every server. Three servers means three independent client connections.

You register each server in the host's config under mcpServers. Add an entry and the host knows one more place to find tools.

At startup the host asks each server for its tools and merges them into a single pool. The model sees one combined list, not three separate ones.

Each server runs in its own process. If the search server crashes, the files and GitHub servers keep working, so one fault does not sink the whole app.

The model never thinks in servers. It just sees a flat list of capabilities and picks the best tool for the task, wherever it lives.

When the model calls a tool, the host knows which server owns it and routes the request to that server's client. The reply travels back the same path.

Because servers are independent, you can add, remove, or upgrade one without touching the others. Your toolset grows like installing plugins. 🧩

A fleet of small servers behind one host gives the model rich, combined powers while keeping each piece simple and replaceable. ✨

How does a host manage three separate MCP servers?

A host runs many small servers, each with its own client, and pools all their tools. Isolation keeps faults contained and lets you swap pieces freely. 🎉