The spread of fake news on social media poses a growing challenge to public trust and online safety. This study introduces a system that detects fake news by combining crowd-sourced data with large language models in an agent-based framework. We initially built a dataset using posts from the Threads platform, which were subsequently verified and labeled through trusted fact-checking sources. To identify real and fake news, a transformer-based classifier was then developed using contextual text embeddings. Furthermore, we implemented an agentic Retrieval-Augmented Generation (RAG) workflow to enhance the accuracy and reliability of the model. This includes tools for web search, source verification and reasoning using a large language model. The final developed system retrieves relevant evidence step by step and checks the credibility of the information. The experimental results show that the system improves detection accuracy and reduces errors while providing clearer explanations for its decisions. Importantly, the approach is adaptable and can handle evolving types of misinformation across platforms. The scrapped dataset used in the agentic workflow and the entire model is made available at https://tinyurl.com/3y8xay2e .