Indexes your codebase locally and exposes search operations to AI coding agents through MCP, cutting input tokens by 94% in benchmarked sessions. Instead of re-reading entire files on every Claude Code or Cursor session, your agent queries a local vector index built with tree-sitter parsing and embeddings (bundled local model or Ollama). Run `cce init` in a project directory and it writes the right config for Claude Code, VS Code, Cursor, Gemini CLI, or Codex CLI, then installs git hooks to keep the index fresh. The context_search tool returns relevant code chunks, and session stats track actual dollar savings based on Sonnet pricing. Everything stays on your machine, no cloud sync involved.
claude mcp add --transport stdio elara-labs-code-context-engine uvx code-context-engine