GitHub

code-lens-aI is an open-source AI developer tool that gives coding agents real code intelligence through an MCP-native server, built for developers who use Claude Desktop, Cursor, Codex, Cline, or any MCP client. Instead of relying on grep plus embeddings, it parses source with Tree-sitter, stores the result in Neo4j, and applies Leiden community detection to surface real architectural subsystems and their spine files. Its 10-tool MCP interface includes get_overview for architectural mapping, impact_analysis that returns a plain verdict such as "Risky: 11 callers, 3 subsystems" instead of a data dump, generate_wiki for a structural wiki skeleton in one call, and call-graph tools like get_callers, get_callees, and get_definition. Search is keyword, semantic, and hybrid. The indexer is incremental and git-aware, re-indexing a 2,900-file repo in around 1.5 seconds. Everything runs locally on Neo4j and Docker, so code never leaves your machine. It is designed for engineers who want AI agents to reason over the actual call graph and subsystem structure of a codebase rather than scattered text matches.

Giving AI coding agents grounded code intelligence via an MCP server backed by a real call graph and architectural subsystems.