Automatic knowledge graph construction is a critical technology for advanced AI applications, traditionally relying on pipelines of tasks such as Named Entity Recognition, Relation Extraction, and Entity Linking. This pipeline approach suffers from system complexity, limited knowledge sharing, and cascading errors. While Universal Information Extraction (UIE) models aim to consolidate these tasks, they critically overlook the essential task of Entity Linking. To bridge this gap, we propose LinkUIE, inspired by Open-Domain Question Answering and Retrieval-Augmented Generation, which seamlessly models Entity Linking within the UIE framework. Our approach first employs a hybrid retrieval module to efficiently retrieve candidate entities from a knowledge base. Subsequently, it innovatively introduces a dynamic sample library. This library adaptively constructs contextual prompts by analyzing the semantic properties of the input text, guiding the UIE model to precisely locate entity mentions. Those tailored prompts enable the UIE model to precisely locate entity mentions and link them within the unified architecture. This framework allows a single model to independently manage the core workflow of knowledge graph construction, significantly reducing complexity and making it ideal for resource-constrained environments without degrading performance on other extraction tasks. Experimental results on the ELEVANT benchmark validate the effectiveness of our framework. For instance, by integrating our adapter, the OneKE model achieves a competitive F1 score of 63.64% on the Kore50 dataset. This result demonstrates the practical viability of our approach and highlights its strong potential to advance fully automated and unified frameworks for knowledge graph construction.