The Model Context Protocol (MCP) is a standard way to connect an AI assistant to external tools and data: your files, a GitHub account, a database, a ticketing system. With it wired up, the model can act - create an issue, read a file, run a query - instead of only talking about those things.
In the tool-calling topic you gave a model an ability by hand: you wrote a schema describing a function, sent it with the request, and ran the loop that handled the model's call and fed the result back. It worked, but every tool was bespoke. You wrote the schema, the glue, and the loop yourself, once per app.
MCP standardizes that whole arrangement. A tool provider writes an MCP server once - it exposes a set of tools and data through the protocol. Any MCP-aware assistant (the client) can then use that server without you writing schemas or a call loop. The assistant and the server already agree on how to describe tools, pass arguments, and return results.
So MCP is not a new capability on top of tool calling. It's the same idea made reusable: a common plug that tool providers and assistants both build to.
Two roles show up every time:
claude CLI is an MCP client.You point the client at a server once. From then on the model can call that server's tools during a normal conversation, the same way it answers any other request.
Because a server is written once and reused, you don't rebuild an integration
for every assistant or every project. Install the GitHub server and both
claude and any other MCP client can work with GitHub. The model moves from
answering questions about your work to doing pieces of it.
Next: how to register a server with claude and where a server's secrets go.
Your lab setup
This VM comes preconfigured with your AI keys - just type claude to get started. Your key and URL are already set as $OPENAI_BASE_URL and $OPENAI_API_KEY.