The multi-contrast Magnetic Resonance Imaging (MRI) super-resolution (SR) seeks to improve MR image resolution by leveraging multiple contrasts. Traditional diffusion-based methods, however, introduce complex conditional constraints, such as LR and multi-contrast MR images, often yielding inconsistent results. Moreover, existing techniques struggle to accurately capture both global and local relationships between the various multi-contrast MR images and their corresponding low-resolution (LR) counterparts. To overcome these limitations, we propose a novel approach: an iterative consistent attentional diffusion model for multi-contrast MRI SR. Our model comprises an iterative diffusion model, which mitigates complex conditional constraints by iteratively separating SR and diffusion process, and a consistent attentional fusion network. The latter effectively marries global and local correlations across multi-contrast and LR MR images through innovative pyramid cross-attention and deformable channel attention mechanisms. We have also developed a dual balance loss function that finely balances denoising with super-resolution enhancement. Experimental results demonstrate the effectiveness of our method in advancing MRI SR.