Computational Grids are powerful platforms gathering computational power and storage space from thousands of geographically distributed resources. The applications running on such platforms need to efficiently and reliably access the various and heterogeneous distributed resources they offer. This can be achieved by using metadata information describing all available resources. It is therefore crucial to provide efficient metadata management architectures and frameworks. In this paper we describe the design of a Grid metadata management service. We focus on a particular use case: the Knowledge Grid architecture which provides high-level Grid services for distributed knowledge discovery applications. Taking advantage of an existing Grid data-sharing service, namely JuxMem, the proposed solution lies at the border between peer-to-peer systems and Web services.