Cortex builds a local knowledge graph of your codebase using tree-sitter parsing, then exposes it to Claude and other MCP clients through semantic search and relationship queries. It indexes files, symbols, architectural decision records, and the connections between them (calls, imports, deprecations), so your assistant retrieves the right context instead of dumping entire directories into the prompt. The setup is one command: `cortex init --bootstrap` scaffolds the index, installs git hooks to keep it fresh, and registers the MCP server with Claude Desktop or Codex. You get graph traversal queries, call graph lookups, and rule enforcement at retrieval time. Runs entirely local with WSL support for Windows users running Node in a Linux environment.
claude mcp add --transport stdio danielblomma-cortex uvx cortex