Module sampling 

Sampling API for server-initiated LLM requests.

This module allows MCP servers to request LLM completions from clients that support the sampling capability. This enables agentic workflows where the server can leverage the client’s LLM for complex tasks.

Example

use mcp_kit::server::sampling::{SamplingRequest, SamplingClient};

async fn agentic_tool(client: impl SamplingClient) -> Result<String, Error> {
    let request = SamplingRequest::new()
        .add_user_message("Analyze this data and suggest improvements")
        .max_tokens(1000);
     
    let response = client.create_message(request).await?;
    Ok(response.content.text().unwrap_or_default())
}

Structs

ChannelSamplingClient

Channel-based sampling client implementation.

NoOpSamplingClient

No-op sampling client for when client doesn’t support sampling.

SamplingRequestBuilder

Builder for creating sampling requests.

Traits

SamplingClient

A client for making sampling requests to the MCP client.