This gives Claude (Desktop, Mobile, and Code) shared access to a self-hosted ChromaDB vector database over HTTP. You get all the standard CRUD operations for collections, embeddings, and semantic search wrapped as MCP tools, plus a REST API proxy for direct Python or JavaScript access. The architecture is straightforward: a Node.js gateway handles auth and protocol translation between MCP clients and your ChromaDB instance. Useful if you want persistent vector memory that syncs across all your Claude clients without round-tripping through Anthropic's servers. Ships as a Docker image with one-command setup, token auth, and Swagger docs at /docs. The main win is keeping your embeddings on your own infrastructure while maintaining the same semantic search context everywhere.
claude mcp add --transport stdio meloncafe-chromadb-remote-mcp uvx chromadb-remote-mcp