Function samplingModelOptions

pkg/runtime/sampling.go:216–225

samplingModelOptions translates the server's advisory preferences into the host's model options. Only MaxTokens is honoured today (with an upper bound enforced by the underlying provider); temperature, stop sequences, and ModelPreferences are intentionally left to the host's configuration. Structure

(req *mcp.CreateMessageParams)

// reshape the model's reply into something the MCP server didn't ask
// for.
func samplingModelOptions(req *mcp.CreateMessageParams) []options.Opt {
	opts := []options.Opt{
		options.WithStructuredOutput(nil),
		options.WithNoThinking(),
	}
	if req.MaxTokens > 0 {
		opts = append(opts, options.WithMaxTokens(req.MaxTokens))
	}
	return opts
}

// drainSamplingStream reads a chat completion stream to completion and
// returns the concatenated assistant content alongside the final finish