Summary
The idea with the meaning of the word sampling is to analyze a small sample. In the context of MCP, this is about delegation and how the server delegates part of its work to the client.
A scenario is usually started by a user, such as authoring a blog post or wanting to create a product in a back office solution. The server ends up creating a sampling request and sends that to the client as part of a task it needs help with. The client is then able to respond to said request using an LLM response.
It should also be stated that the sample request contains recommendations on model, token usage, system prompt, and much more, and that it should involve a human who either accepts these recommendations or changes them to their liking.
This is a great functionality, where an LLM on the client can be called in to help.
In our next chapter, we’ll dive into yet another powerful feature of MCP, namely, elicitation, which is about improving the user experience by...
