Sequential Questioning MCP Server

# Automatic Follow-Up Questions This document explains how to use the automatic follow-up questions feature in the Sequential Questioning MCP Server. ## Overview The Sequential Questioning MCP Server now provides three different endpoints for generating questions: 1. **Basic Question Generation** - `/mcp-internal/question` 2. **Follow-Up Question Generation** - `/mcp-internal/question/follow-up` 3. **Automatic Multi-Round Questioning** - `/mcp-internal/question/automatic` These endpoints allow for different levels of automation in the question generation process, from simple single-round questioning to fully automatic multi-round questioning with follow-ups. ## Endpoint Comparison | Feature | Basic Question | Follow-Up Question | Automatic Multi-Round | | --- | --- | --- | --- | | Operation ID | `sequential_questioning` | `sequential_questioning_follow_up` | `sequential_questioning_automatic` | | Designed for | Initial questions | Explicit follow-up questions | Complete conversation flows | | Requires conversation_id | No | No (recommended but optional) | No | | Max question rounds | 1 | 1 | Configurable (1-3) | ### Follow-Up Question Generation The follow-up question endpoint is now more flexible for use as an LLM tool. When a follow-up question is requested: 1. If a `conversation_id` is provided, it will use that existing conversation 2. If no `conversation_id` is provided but `previous_messages` are included, it will: - Create a new conversation automatically - Return the new `conversation_id` in the response - Generate appropriate follow-up questions based on the previous messages This makes it easier for LLMs to use the follow-up endpoint without having to explicitly track conversation IDs. ## Automatic Multi-Round Questioning The `/mcp-internal/question/automatic` endpoint is designed to streamline the questioning process by generating multiple rounds of questions in a single request. This is particularly useful for gathering comprehensive information from users without requiring multiple separate calls to the API. ### Request Format ```json { "context": "Month planning for software developers", "user_id": "user-123", "previous_messages": [ { "role": "user", "content": "I need help planning my workload for the next month with multiple ongoing projects." } ], "auto_handle_follow_up": true, "max_rounds": 2 } ``` ### Parameters - `context` (string, optional): Provides context for question generation - `user_id` (string, optional): User identifier for tracking conversations - `previous_messages` (array, optional): Previous messages in the conversation - `auto_handle_follow_up` (boolean, default: true): Whether to automatically generate follow-up questions - `max_rounds` (integer, default: 2): Maximum number of question rounds to generate (capped at 3) ### Response Format ```json { "initial_questions": { "current_question": "...", "questions": [ { "question_text": "...", "question_number": 1, "metadata": { ... } }, ... ], "conversation_id": "...", "session_id": "...", "current_question_number": 1, "total_questions_in_batch": 5, "total_questions_estimated": 10, "next_batch_needed": true, "metadata": { ... } }, "follow_up_questions": [ { "current_question": "...", "questions": [ ... ], ... } ], "all_questions_combined": "1. First question\n2. Second question\n...", "conversation_id": "...", "session_id": "...", "total_questions": 8, "metadata": { "rounds_generated": 2, "timestamp": "2025-05-15T12:34:56.789012", "auto_follow_up": true } } ``` ### Displaying Questions to Users The `all_questions_combined` field contains a formatted string with all questions from all rounds, numbered sequentially. This makes it easy to present all questions to the user at once. ## Example Usage with cURL ```bash curl -X POST "http://localhost:8000/mcp-internal/question/automatic" \ -H "Content-Type: application/json" \ -d '{ "context": "Month planning for software developers", "user_id": "test-user", "previous_messages": [ { "role": "user", "content": "I need help planning my workload for the next month with multiple ongoing projects." } ], "max_rounds": 2 }' ``` ## Example Usage in an LLM Tool ```python def sequential_questioning_tool(context, user_id=None, conversation_id=None, previous_messages=None, max_rounds=2): """ MCP tool for generating multi-round sequential questions. Args: context: The context for question generation user_id: Optional user identifier conversation_id: Optional conversation identifier for follow-up questions previous_messages: List of previous messages in the conversation max_rounds: Maximum number of question rounds to generate Returns: A formatted string with numbered questions """ url = "http://localhost:8000/mcp-internal/question/automatic" payload = { "context": context, "user_id": user_id, "conversation_id": conversation_id, "previous_messages": previous_messages, "max_rounds": max_rounds } response = requests.post(url, json=payload) response.raise_for_status() data = response.json() return data["all_questions_combined"] ``` ## Using Follow-Up Questions in LLM Tools When integrating the follow-up endpoint as an LLM tool, you can now use it without requiring explicit conversation tracking: ```python def follow_up_question_tool(context, previous_messages, conversation_id=None): """ Tool for generating follow-up questions based on previous messages. Args: context: Conversational context previous_messages: List of previous messages containing user answers conversation_id: Optional conversation identifier (will be created if not provided) Returns: A formatted list of follow-up questions """ url = "http://localhost:8000/mcp-internal/question/follow-up" payload = { "context": context, "previous_messages": previous_messages, "conversation_id": conversation_id } response = requests.post(url, json=payload) response.raise_for_status() data = response.json() # Store the conversation_id for future calls if needed conversation_id = data["conversation_id"] return data["current_question"] ``` ## Best Practices 1. **Balance Depth vs. Length** - While more rounds provide more comprehensive questioning, they can also lead to lengthy exchanges. Consider your user experience when setting `max_rounds`. 2. **Provide Rich Context** - The quality of generated questions improves with better context. Include relevant details about the topic and user needs. 3. **Include Previous Answers** - When using the automatic questioning feature, including previous user answers in `previous_messages` helps generate more relevant follow-up questions. 4. **Handle Long Responses** - When presenting multiple questions to users, ensure your UI can handle potentially lengthy responses. 5. **Conversation Continuity** - Store the `conversation_id` and `session_id` from responses to maintain conversation context for future interactions. While the follow-up endpoint can now work without a conversation_id, providing one when available will result in better context management. 6. **LLM Integration** - When integrating with LLMs, focus on providing rich context and previous messages. The endpoints are now designed to handle conversation tracking automatically if needed.