cortex-local 2.1.0

Local semantic search MCP server — notes + code, embedded Qdrant, fastembed

cortex-local

Local semantic search MCP server for Claude Code — index your notes and code, search them semantically without any server or cloud dependency.

Install

pipx install cortex-local
cortex install

cortex install downloads embedding models and registers the MCP server in Claude Code. Restart Claude Code after.

Legacy CPUs (no SSE4.2 / X86_V2)

NumPy 2.x requires SSE4.2. If you get a RuntimeError: NumPy was built with baseline optimizations (X86_V2) error, use the legacy extra with Python 3.12:

pipx install "cortex-local[legacy]" --python /path/to/python3.12
cortex install

[legacy] pins numpy==1.26.4 which has no SSE4.2 requirement. Python 3.12 is required because numpy 1.26.x has no wheels for 3.13+.

Usage

cortex index ~/notes          # index Obsidian vault
cortex index ~/projects/myapp # index a code repo

Then in Claude Code, search_notes and search_code are available as MCP tools.

How it works

• Embeddings: fastembed (nomic-embed-text-v1.5) + BM25 sparse — runs fully local, no GPU needed
• Vector store: embedded Qdrant at ~/.cortex/qdrant
• Graph layer: import graph with degree centrality re-ranking + Personalized PageRank for notes

Full documentation: github.com/Xoudusz/cortex