BGI analyzes large codebases by grouping code units by behavioral role rather than just syntax references. It exposes MCP tools like task_fingerprint, behavioral_twins, and twin_context that help ground AI prompts in actual repository behavior patterns. The server uses COV tokens to classify behavior, DRS clustering with hard size caps to prevent giant merged clusters, and emits structured artifacts like bgi-graph.json and fuse-graph.json that show architectural boundaries and coupling seams. Reach for this when you need architecture-aware context for refactoring decisions or want to reduce hallucination risk by anchoring AI changes to proven patterns in your codebase. Supports Python, TypeScript, Go, Rust, Java and others with varying parser quality tiers.
claude mcp add --transport stdio ahmedxuhri-bigindexer uvx bigindexer