Attributes of clinical concepts, such as severity and negation modifiers of clinical problems, are vital elements for clinical Natural Language Processing (NLP). Here we introduce CLAMP-ATR (Attribute), a newly developed comprehensive pipeline for attribute recognition with the CLAMP NLP toolkit. It not only achieved high-performance on attribute recognition across different corpora, but also made customization easier by providing a Docker container for training deep learning models and a GUI interface for building pipelines. CLAMP-ATR currently recognizes three types of important clinical concepts and their attributes: 1) Problem and modifiers of subject, negation, severity, body location, etc.; 2) Lab test and its value; 3) Medication and signatures of dose, form, route, frequency, etc. We used the Bi-directional Long-Short-Term Memory (biLSTM) network for both entity and attribute recognitions. Character embeddings and word embeddings were used as input. For attribute recognition, an additional embedding layer was added to the input to represent the primary entities. To achieve the most optimized performance, we developed three attribute models separately. For easier deployment, we built a Docker container for training deep learning models using local data. It works as an insulated virtual environment with all deep learning libraries and parameters pre-set. Users can run this container on their own corpus with just simple docker commands. In addition, the trainer can use GPUs if they are available on the host machine. We believe a high-performance, easy-to-use toolkit to automatically recognize concept and attribute information from clinical text will greatly facilitate related clinical and translational research and applications.