ripwireMCP Server

ripwire describes itself as "the ripgrep of AI context", and the comparison is about shape as much as speed: one self-contained binary, no daemon, no database, no network. It answers the questions a coding agent actually asks before making a change — where is this defined, what calls it, what breaks if I touch it — and it answers them from a fresh index each time rather than from a service you have to keep running.

The performance claims are unusually well documented. Indexing takes 0.25–0.45 seconds in 6.6–16.5 MB, measured against 23–52 seconds and 391–623 MB for the graph-database MCP server it benchmarks itself against, with warm queries at 197 ms against 1,082 ms across 48 matched questions on django, webpack and its own repository. The method, and the losses as well as the wins, are published in docs/EVALS.md. Answers are designed to be terminal — complete enough that the agent does not follow up with three more greps — and land at roughly 5% of the tokens a naive grep-and-read costs.

Beyond lookup, it scores code quality across six independent evidence families: complexity, naming, idiom use, git history, colocation and state side effects. Query modes cover symbol lookup, conceptual search, blast radius and task orientation (--for="<the change you are about to make>"), and an HTML graph view colours the result by complexity, churn or test coverage. Twenty-four vendored grammars span Rust, C++, Python, Go, JavaScript, Java, TypeScript, Kotlin, Dart, Elixir and Ruby.

Worth noting how the maintainers position the MCP server: the CLI is the cheaper interface and the recommended first stop, because MCP verb schemas sit in the agent's context every session whether or not they are called. The installer also activates task-shaped skills for every agent it finds — Claude Code, Codex, Cursor, Windsurf, Gemini, opencode and aider. Maintained by Red Hat Emerging Technologies under Apache-2.0. Requires CMake 3.24+ and a C++23 compiler if building from source.