The author presents a distributed constraint processing approach for planning flow production plants. It is based on a combination of constraint and multi-agent technology. The approach is realized in a production planning system, called FAKOS. It was developed for a car factory of Mercedes-Benz. System analysis and design is based on multi-agent technology. This allows an intuitive representation of knowledge about plants. Properties of plants and interconnections between them are represented as constraints. The planning process of FAKOS is based on dynamic generation and processing of partial constraint networks. The control of the problem solving process is described in form of a multi-agent coordination protocol. This protocol employs a new approach for intelligent backtracking. Dependencies between partial solutions are handled by communication of value assignments and constraints. In case of a conflict, additional constraints are generated pruning the search space better than existing intelligent backtracking methods. This article describes the constraint modeling and partitioning, the control mechanism, the constraint processing and the generation of the additional constraints.