THE MODEL CONTEXT PROTOCOL (MCP) AND TOOL STANDARDS

WHAT SERVERS OFFER An MCP server can offer several kinds of capabilities. Tools are functions the model can ask to run, just as in Chapters 13 to 15. Resources are pieces of data the host can read, such as a file's content or a database record, meant to supply context. Prompts are reusable templates for common interactions, which a user can pick. The specification also describes features that clients can offer back to servers, such as asking the host's model to generate text on the server's behalf (called sampling), asking the user a question (elicitation), and telling a server which folders it may look at (roots). Because the specification is revised over time, details such as which features are stable and how authorization works can change, so always consult the current official documentation for the version you use. When the client first connects, it and the server perform a handshake in which they greet each other and agree on the version and the features both support. Then the client asks the server to list its tools, receives names, descriptions, and input schemas, and passes them to the model. When the model requests a tool, the host sends the call to the right server, receives the result, and returns it to the model. This fits perfectly into the harness picture you already know. MCP tools simply arrive in the registry from outside instead of being written in your own code.