MCP Scout (smarter-faster-better-mcp)

@iHaiduk

MCP Scout (smarter-faster-better-mcp) について

A blazing fast, ultra-lightweight Model Context Protocol (MCP) server for AST-based code intelligence and semantic discovery. It runs as compiled Node.js JavaScript for npx, uses oxc-parser by default for JavaScript/TypeScript, and can opt into web-tree-sitter for polyglot reposi

設定

以下の設定を使って、このサーバーを MCP 対応クライアントに追加してください。

{
  "mcpServers": {
    "scout": {
      "command": "npx",
      "args": [
        "smarter-faster-better-mcp"
      ]
    }
  }
}

ツール

ツールは検出されませんでした

ツールは README から自動的に抽出されます。メンテナーは ## Tools という見出しの下に記載することで、このタブに反映できます。

概要

What is MCP Scout?

MCP Scout is a blazing fast, ultra-lightweight Model Context Protocol server for AST-based code intelligence and semantic discovery. It uses the Rust-powered oxc-parser by default for JavaScript/TypeScript and can optionally use web-tree-sitter for polyglot repositories (Python, Go, Rust, Java, C/C++, C#, PHP, Dart, Ruby).

How to use MCP Scout?

Run directly via npx smarter-faster-better-mcp. Configure an LLM provider by creating a .env file in your project root with SCOUT_BASE_URL, SCOUT_API_KEY, and SCOUT_MODEL. Add the server to your MCP client (Claude Desktop, Cursor, Windsurf, Trae, Claude Code) using the command npx smarter-faster-better-mcp. Optionally set the --parser tree-sitter flag for polyglot support.

Key features of MCP Scout

Zero native bindings — installs in milliseconds without node-gyp
Deterministic Jaro-Winkler filtering for instant symbol ranking
LLM chunk parallelism distributes classification requests efficiently
AST extraction with summaryOnly mode saves up to 90% of tokens
AST + regex dependency mapping via ripgrep
Read-only contract — never modifies project source files
Supports JavaScript, TypeScript, Python, Go, Rust, Ruby, Java, C/C++, C#, PHP, Dart

Use cases of MCP Scout

Rapid code search and symbol discovery in large JavaScript/TypeScript projects
Tracing exact dependency imports across files in a monorepo
Providing LLMs with summarized file context without reading entire raw files
Analyzing polyglot codebases (e.g., React frontend + Python backend)
Reducing token consumption and context overflow during AI-assisted code analysis

FAQ from MCP Scout

What are the runtime requirements for MCP Scout?

Node.js >= 18, Bun >= 1.1.0 (for development testing), and Ripgrep (rg) for symbol dependency lookup.

How does MCP Scout achieve such high speed?

It uses the Rust-based oxc-parser for instant file parsing, deterministic Jaro-Winkler edit distance filtering, LLM chunk parallelism, and AST extraction that collapses code bodies into stubs, saving tokens.

Does MCP Scout modify my project files?

No, it is strictly read-only. The only writes are to its own cache files (.project_map.json and .scout-cache/), which can be safely cleaned with the cleanup_workspace tool.

Which parser engine should I choose?

Use the default oxc for pure JavaScript/TypeScript codebases. Switch to tree-sitter (via --parser tree-sitter) for monorepos or backend services that include Python, Go, Rust, Ruby, Java, C/C++, C#, PHP, or Dart files. Even in tree-sitter mode, JS/TS files are still parsed by oxc for maximum performance.

How do I configure MCP Scout?

Create a .env file in your project root with the required variables SCOUT_BASE_URL, SCOUT_API_KEY, and SCOUT_MODEL. Optional variables include SCOUT_LLM_TIMEOUT_MS, SCOUT_CONFIDENCE_THRESHOLD, SCOUT_LLM_PARALLELISM, and SCOUT_PARSER. The server loads these automatically on startup.

よくある質問

What are the runtime requirements for MCP Scout?

Node.js >= 18, Bun >= 1.1.0 (for development testing), and Ripgrep (`rg`) for symbol dependency lookup.

How does MCP Scout achieve such high speed?

Does MCP Scout modify my project files?

No, it is strictly read-only. The only writes are to its own cache files (`.project_map.json` and `.scout-cache/`), which can be safely cleaned with the `cleanup_workspace` tool.

Which parser engine should I choose?

Use the default `oxc` for pure JavaScript/TypeScript codebases. Switch to `tree-sitter` (via `--parser tree-sitter`) for monorepos or backend services that include Python, Go, Rust, Ruby, Java, C/C++, C#, PHP, or Dart files. Even in tree-sitter mode, JS/TS files are still parsed by oxc for maximum performance.

How do I configure MCP Scout?

Create a `.env` file in your project root with the required variables `SCOUT_BASE_URL`, `SCOUT_API_KEY`, and `SCOUT_MODEL`. Optional variables include `SCOUT_LLM_TIMEOUT_MS`, `SCOUT_CONFIDENCE_THRESHOLD`, `SCOUT_LLM_PARALLELISM`, and `SCOUT_PARSER`. The server loads these automatically on startup.