Skip to content

v0.3.1 — a tag whose tree matches what master says

Latest

Choose a tag to compare

@satovarb16 satovarb16 released this 07 Sep 20:05
· 7 commits to master since this release

A packaging-only release. No server code changed since v0.3.0server.py, tools/, and every tool contract are byte-identical. If the plugin is already running for you, nothing about its behavior changes.

Why this exists

The v0.3.0 tag pointed at a commit three behind master. Its tree carried an outdated README and a plugin pin of runway-mcp==0.3.0 — a version that does not exist on PyPI, which still serves 0.1.2. So anyone browsing or installing from that tag got stale docs and a pin that fails outright.

The tag could not simply be moved: uvx resolves a git ref once and caches the commit it found, so a moved tag leaves everyone who already installed it building the old commit forever. Hence a new version.

v0.3.1 is the first tag whose tree actually carries the working git pin.

Install

/plugin marketplace add satovarb16/runwayMCP
/plugin install runway-mcp@satovarb

Or manually, in a .mcp.json where you run Claude Code:

{
  "mcpServers": {
    "runway-mcp": {
      "command": "uvx",
      "args": ["--from", "git+https://github.com/satovarb16/runwayMCP@v0.3.1", "runway-mcp"]
    }
  }
}

Already on the plugin? /plugin marketplace update satovarb then /plugin update runway-mcp@satovarb, and /reload-plugins.

The git URL is deliberate. Plain uvx runway-mcp pulls 0.1.2 from PyPI — a much older server from before the SQLite rewrite, with a tool set that no longer matches the README. PyPI publishing is blocked pending account recovery.

Release-process fixes

  • The version checklist said "four places". There are six. It omitted manifest.json's --from pin, the README's manual-install snippet, and the test asserting the version. The README omission was the dangerous one — nothing checked the snippet every manual installer copy-pastes.
  • The version guard no longer hardcodes a version. test_version_is_0_3_0_everywhere had the number baked into its own name, so every release had to edit the test meant to be guarding it. It now reads the version from pyproject.toml and asserts the other five sites agree.
  • The publish job is gated on a repository variable instead of failing on every tag with invalid-publisher. This is the first Release run in this repo's history to come back green.

Full changelog: v0.3.0...v0.3.1