Abstract: This study develops a large language model-based agent framework for the finite element analysis of truss and framed structures, integrating natural language parsing, finite element modeling, code generation and execution, and result verification into a unified workflow. Numerical examples demonstrate that the framework can extract the information and parameters required for finite element modeling from natural language descriptions and automatically generate and execute finite element analysis code. Compared with reference solutions obtained using commercial finite element software, the relative errors in nodal displacements and member internal forces are all less than 0.01%, demonstrating the high computational accuracy of the proposed framework. The framework provides intelligent assistance for users with limited expertise in finite element analysis to perform mechanical analysis of framed structures.