Multi-view clustering has received significant interest because different views contain various features that provide a better description of the target object. However, most of the current multi-view clustering methods only consider the global and local information of the graph, thus ignoring the global and local clustering structure of pseudo label. In addition, some methods ignore redundant features or features with unclear clustering structures that may be present in the original data. To this end, a method named Feature-guided Multi-view Clustering by jointing Local subspace label learning and Global label learning (MCLG) is proposed. Specifically, local and global label learning are integrated into one framework. Based on the anchor strategy, we propose to learn a unified local pseudo label for all views in the subspace. In the original space, the global graph is taken into a unified graph fusion learning strategy, making the learned global pseudo label contain richer information. In addition, the feature-guided learning strategy is added to the framework to select valuable features that can facilitate the learning of the clustering structure. Then, through the subspace graph structure reconstruction learning strategy, the subspace graph structure information is effectively utilized, and the integrity of the graph structure information is guaranteed. Finally, an algorithm that effectively guarantees convergence is proposed to optimize the proposed algorithm, and the effectiveness of MCLG is verified by the results obtained from nine real-world datasets.