We implement an LLM-based coaching agent using motivational interviewing (MI) codes to assess users’ motivational states and select appropriate coaching strategies. This approach balances rapport-building and goal-seeking strategies, acknowledging that MI’s effectiveness depends on a client’s readiness for change. The study will compare non-MI and MI-aware conditions, as well as text-only and embodied virtual agents. Using standardized psychology questionnaires, we evaluate agent competency and human-agent therapeutic alliance, examining this connection’s role in successful coaching. This alliance was voted among the top 10 research priorities in a survey involving over 600 mental health stakeholders in the UK.