north-rune 0.1.0

The Software Intelligence Runtime — persistent, explainable software understanding for any AI. Python distribution.

Rune (Python distribution)

The same Software Intelligence Runtime as the JS/TS package, packaged for pip install — for teams whose tooling lives in Python (CI, monorepo scripts, internal tooling) who'd rather not require Node.js just to run Rune against a JS/TS/React/Next.js/Express codebase.

Feature parity with the JS version in v0.1: same fact/derived-understanding model, same React/Next.js/Express detection, same MCP tool surface, same documented limitations (heuristic regex-based extraction, no cross-file Express route-prefix resolution, dotfiles/secrets never scanned).

Package name

Published on PyPI as runesdk (the bare rune name is contested/likely taken — a crowded cluster of rune-* packages already exists). The CLI command itself is still rune — only the package name you pip install differs, same as the JS side installs as @north-research/rune but runs as rune.

Install

pip install runesdk

For local development from source:

cd python
pip install -e .

Quickstart

cd your-project
rune init
rune scan
rune serve

Same CLI shape as the JS version: rune init [dir], rune scan [dir], rune serve [dir], rune explain <id>, rune --version.

Requirements

• Python 3.10+
• The mcp package (installed automatically as a dependency) for rune serve

Testing

No pytest dependency required — the suite runs on Python's built-in unittest:

python -m unittest tests.test_scan -v

To verify the live MCP server (the one thing the unit tests can't cover, since it needs the real mcp package installed and a real subprocess handshake):

pip install -e .
python scripts/verify_mcp_server.py
# or against a real project:
python scripts/verify_mcp_server.py /path/to/your-project

Differences from the JS version worth knowing

• The Python mcp SDK's tools are defined as explicit @mcp.tool()-decorated functions (src/rune/mcp/server.py), not a dynamically-built list like the JS side's tools.js. This was a deliberate choice: the decorator pattern is the most version-stable, best-documented part of the Python MCP SDK's API, which matters most for the one part of this port that couldn't be tested without a real install.
• ID formatting differs cosmetically (hex counters, same scheme) but the scoping guarantee is identical: ids reset per scan and are unique within a scan.