Working with Prompts

The prompt system in MCP provides a standardized way to define, store, and retrieve prompt templates for use with Large Language Models (LLMs). Prompts in MCP are predefined message sequences that can be either static (with fixed content) or templated (with customizable parameters).

Listing Prompts

To list available prompts:

List<PromptItem> prompts = mcpClient.listPrompts();
assertEquals(5, prompts.size(), "Expected 5 prompts");

PromptItem basic = findPromptByName("basic", prompts);
assertNotNull(basic, "Prompt 'basic' should not be null");
assertEquals("Basic simple prompt", basic.getDescription(), "Description mismatch for 'basic'");
assertNull(basic.getArguments(), "'basic' arguments should be empty");

static PromptItem findPromptByName(String name, List<PromptItem> promptRefs) {
        for (PromptItem promptRef : promptRefs) {
            if (promptRef.getName().equals(name))
                return promptRef;
        }

    return null;
}

Getting Prompt Details

To get details of a specific prompt:

GetPromptResult.PromptResultDetail prompt = mcpClient.getPrompt("basic", new HashMap<>());
assertEquals("Basic simple prompt", prompt.getDescription(), "Description should not be null");
assertEquals(1, prompt.getMessages().size(), "Expected exactly one message");

PromptMessage message = prompt.getMessages().get(0);
assertEquals(Role.USER, message.getRole(), "Role should be USER");
assertEquals("Hello", ((ContentText) message.getContent()).getText(), "Text content should be 'Hello'");

The getPrompt() method returns a PromptResultDetail containing the prompt's description and messages.

Normally, the prompt system receives a prompt name and a map of arguments. The arguments are used to fill in the placeholders in the prompt template. It accepts a map of arguments, where the key is the argument name and the value is the argument value. And JSON String also accepts.

If you don't want to use arguments, you can pass an empty map.