scc-mcp

An MCP server that exposes boyter/scc — a fast code-counting and complexity tool — to LLM agents.

It wraps scc in a small set of read-only tools so an agent can count lines of code, find the largest or most complex files, break stats down by language, and estimate project cost without learning the scc CLI.

Requirements

• uv
• scc installed and on your PATH (it is an external binary, not a Python dependency, and is not installed for you). Verify with scc --version.

Install

uv sync

Run

The server speaks MCP over stdio:

uv run scc-mcp

Tools

All tools are read-only. They default to scanning the current working directory and return human-readable markdown, or structured JSON when response_format: "json" is passed. Most accept include/exclude extension and directory filters.

Tool	Description
scc_count_lines	CLOC (code/comment/blank/lines) per language, plus totals.
scc_top_files	Top N files sorted by a metric (code, complexity, lines, ...).
scc_by_extension	CLOC broken down per language/extension.
scc_complexity_hotspots	Files ranked by cyclomatic complexity (refactor targets).
scc_cost_estimate	COCOMO cost / schedule / people estimate.
scc_file_metrics	Detailed per-file metrics for a file or filtered subset.
scc_list_languages	Languages scc recognizes and their file extensions.

Use with opencode

This repo registers the server as a local MCP in opencode.json:

{
  "mcp": {
    "scc": {
      "type": "local",
      "command": ["uv", "run", "scc-mcp"],
      "enabled": true
    }
  }
}

For other MCP clients, point them at the same uv run scc-mcp command (stdio transport).

Development

uv run pytest     # run the test suite
make build        # build sdist + wheel into dist/

Tests that shell out to scc are skipped automatically when it is not on PATH. See AGENTS.md for module layout and conventions.