What is scopewalker-mcp?

Scopewalker MCP is a tool that helps AI coding assistants keep code quality in check. AI agents will often create huge files or add too many parameters to a function even when told not to, so Scopewalker gives the assistant a way to actually check the codebase and enforce stricter standards instead of just following instructions blindly. It works as a local MCP server, a kind of plugin format that AI coding assistants like Claude Code, Cursor, VS Code, Windsurf, Gemini CLI, and OpenAI Codex CLI can connect to. It runs entirely on your machine over standard input and output, makes no network calls, and offers eight read only tools the assistant can call. These tools count lines of code per file, detect functions and measure their length, calculate complexity scores like nesting depth and parameter counts, flag files or functions that exceed size limits, list out classes, functions, and other code structures, measure how much of the code has documentation comments, find markers like TODO or FIXME left in comments, and detect when the same parameter name gets passed through many functions and files, a pattern known as prop drilling. Under the hood it relies on existing tools rather than writing its own parser: tree-sitter for reading code structure, tokei for counting lines, and fast-glob for finding files. It has been tested on macOS with Claude Code but should work with any tool that speaks the MCP protocol. For safety, it only looks inside allowed folders, which default to your current working directory and the system temp folder, skips very large files to avoid slowing down your machine, and limits how much output it returns unless you ask for more. To use it you need Node.js 22 or newer and tokei installed separately, then you can install the package from npm or configure your AI assistant to run it directly with npx, no manual build required unless you want to develop it further. It supports analyzing TypeScript, JavaScript, Python, Go, Rust, Java, C, C++, and Ruby code, and is released under the MIT license.