Collective Brain

Dockerized Model Context Protocol server uniting IDEs and local agents over Qdrant semantic retrieval and cross-encoder reranking with zero GPU VRAM allocation.

Built to unify fragmented context across terminal sessions, editors, and web tools without starving co-hosted services. Runs two-stage retrieval — FastEmbed ONNX embeddings (BAAI/bge-small-en-v1.5) into Qdrant followed by cross-encoder reranking (ms-marco-MiniLM-L-6-v2) — constrained strictly to CPU with single-threaded ONNX execution, query timeouts, and markdown-aware chunking. Exposes 18 tools across Streamable HTTP, legacy SSE, and OpenAPI 3.1.0 REST.