Lain

Builds a persistent knowledge graph of your codebase using LSP and tree-sitter, then exposes it through MCP tools so agents can answer structural questions like "what breaks if I change this function?" Runs as a background server with stdio or HTTP transport, watches files to keep the graph fresh, and layers git co-change history on top of the static analysis. Exposes operations like query_graph for traversals, get_blast_radius for impact analysis, find_anchors to identify architectural pillars, and semantic_search if you point it at a local ONNX embedding model. Also wraps build commands (run_build, run_tests, run_clippy) and decorates failures with caller context. Reach for this when your agent needs to reason about cross-file dependencies and architectural structure, not just grep through isolated files.