Multi-day travel itinerary planning is increasingly vital for personalised tourism, yet it poses significant challenges due to complex constraints like budget, time, and user preferences. Traditional planning methods, unable to process natural language directly, often fail to capture user intent accurately. While Large Language Models (LLMs) enable natural language understanding, single-agent systems built upon them frequently exhibit hallucinations, leading to logical inconsistencies and poor constraint adherence in complex scenarios. To address these issues, we propose the Multi-round Feedback Travel Planner (MFTP), an intelligent multi-agent framework that enhances planning through iterative feedback and cooperative optimisation. MFTP integrates intent parsing, planning and evaluation agents. The parsing module structures user requirements, the planning agent drafts an initial itinerary, and the evaluation agent validates constraints across multiple dimensions, with iterative interplay refining the plan. Experiments on the TravelPlanner dataset demonstrate MFTP’s superiority over baseline methods, significantly enhancing pass rates and constraint satisfaction, and effectively resolving logical conflicts and constraint violations in multi-day travel planning. Code can be found at https://github.com/lsnuNLP/ems