mcp-cpp

C++ codebases are hostile territory for text-only browsing — preprocessor macros expand, templates instantiate into types that don’t appear anywhere in the source, and inheritance hierarchies span files a grep will never connect. mcp-cpp closes that gap by driving clangd, the same engine behind IDE IntelliSense, and exposing its understanding through three MCP tools: get_project_details for build-environment discovery, search_symbols for boundary-aware symbol search, and analyze_symbol_context for inheritance and call-hierarchy analysis.

It’s built for large, multi-component codebases specifically — the kind found in embedded Linux projects — and can track several C++ components simultaneously, auto-detecting CMake or Meson build directories (or switching between them) and distinguishing project code from external dependencies so search results aren’t drowned in library internals. An indexing monitor watches clangd’s index state directly to confirm full symbol coverage before results are trusted.

Written in Rust, it ships as a cargo install-able binary or a Docker image bundling clangd-20 on an Ubuntu base. Either way, the target project needs a compile_commands.json — generated by CMake or Meson — since that’s what clangd itself requires to reason about the code.