Conversational Recommender Systems (CRSs) aim to deliver personalized guidance via iterative dialogue exchanges. While Large Language Models (LLMs) have achieved convincing performance on conversational and recommendation tasks concurrently, integrating user preference, contextual knowledge, and generation quality remains a significant challenge. Therefore, we propose GraphPromptCRS, a prompt-based and knowledge-grounded CRS framework that jointly performs recommendation and response generation with a frozen LLM. Our system leverages soft prompt learning to encode task-specific information without fine-tuning all model parameters. To enhance the reasoning capabilities, we introduce a GraphRAG-based knowledge construction pipeline that builds knowledge graphs from dialogue history using structured prompts. Additionally, we incorporate a Community Prompt Enhancer to capture users’ topical preferences, guiding personalized and context-aware generation. Extensive experiments conducted on the ReDial dataset indicate that GraphPromptCRS significantly outperforms baselines in recommendation accuracy and conversational diversity, highlighting the performance of our approach.