Hi everyone, I'm the maker of LayerMap. I built it around a question that comes up before changing existing code: “If I modify this function, which endpoints or jobs could be affected?” LayerMap gives coding agents a local code map through three read-only MCP tools. It uses language compilers and type checkers to resolve references and trace callers through multiple levels. The README includes a Claude Code demo on Miniflux: one map query finds four callers of Storage.MarkFeedAsRead and follows their upstream entry points. The agent then checks the relevant source. It supports TypeScript, JavaScript, Go, Python, and Java, and works with Claude Code, Codex, and other MCP clients. Everything runs locally, with no source code uploaded. A few limitations: it currently supports macOS and Linux, the first analysis takes time, and dynamic calls or reflection can still be missed. It's open source under Apache-2.0. I'd love feedback from people trying it on their own codebases: Was setup straightforward? Did it miss an important relationship? Was the output useful for deciding what to inspect next?

LayerMap

Local code maps for AI coding agents

Local code maps for AI coding agents