repolayer: Instructions file for Claude Code

CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

Project

repolayer is a Rust CLI + MCP server that pre-computes 4 SQLite-backed indices (graph, per-file outline, file-level deps, hybrid search) across one or more repos and serves them to AI agents via MCP. Single static binary, ast-grep-core parsing for source languages, bare tree-sitter for IDL (protobuf/thrift). No runtime deps for end users.

Status: v0.3.0. Repositioned as the cross-repo + IDL extension of aeroxy/ast-outline — see NOTICE for adopted components.

Common commands

# Build / test
cargo build                                  # debug build
cargo build --release                        # release (LTO, stripped, ~47 MB)
cargo test                                   # run all 290+ integration tests
cargo test --test cli_build                  # run a single integration test file
cargo test adapter_                          # all adapter tests by name prefix
cargo test -- --nocapture                    # show test stdout/stderr
cargo clippy --all-targets -- -D warnings    # lint (must be clean)
cargo fmt                                    # format

# End-to-end usage
cargo run -- init                            # creates repolayer.yml in CWD
cargo run -- build                           # full index → 4 SQLite files in .repolayer/
cargo run -- update                          # incremental re-index via `git diff`
cargo run -- serve                           # start MCP stdio server (logs to stderr)
cargo run -- install --mcp claude-code       # write claude_desktop_config.json
cargo run -- prompt                          # output agent-steering snippet

serve uses stdio for MCP — stdout is reserved for the protocol, all tracing output goes to stderr (configured in src/main.rs). Don't add println! to anything reachable from the serve path.

Architecture

Pipeline (build): config → walk repos in parallel via rayon → ast-grep parse (or bare tree-sitter for IDL) → mpsc-channel-style serial write to 4 SQLite stores → cross-repo gluing → search index → optional LLM summary.