Claude Context: MCP setup with Milvus and Ollama

Claude Context is presented as an open source MCP server that gives a coding assistant semantic search over a repository. It indexes code in a vector database and retrieves relevant chunks for questions about authentication, retry logic or other implementation details. The speaker describes hybrid retrieval using dense vector search and BM25 keyword search, plus AST-based chunking with a fallback splitter. Incremental indexing uses Merkle trees to update changed files.

The setup overview covers Zilliz Cloud with OpenAI embeddings and API keys. The stated Node requirement is version 20 or above but below 24; the speaker says Node 24 is unsupported. For a self-hosted retrieval setup, the video cites the project's FAQ: use Milvus locally and Ollama for embeddings. This describes the search infrastructure, rather than establishing that every connected AI agent runs locally.

The workflow has four operations: index a codebase, search code, check indexing status and clear an index. The speaker recommends it mainly for medium to large repositories and notes that small projects may need only ordinary search. Maintaining a vector database adds work, and retrieved code still needs model reasoning. Claude Context is also described as distinct from persistent session memory. The cited claim of roughly 40% fewer tokens comes from the project's own evaluation, not an independent test.