skyboy-mcp (PyPI)

A thin PyPI wrapper for the Skyboy MCP server. The server itself is implemented once in TypeScript (@skyboy/mcp-server) and published on npm. This package runs that same core for Python users: it checks that Node.js is present, then hands off to npx -y @skyboy/mcp-server.

Install

pipx install skyboy-mcp
# or
pip install skyboy-mcp

Usage

skyboy-mcp

This starts the stdio MCP server locally. Point an MCP-compatible agent at it with the stdio transport:

{
  "mcpServers": {
    "skyboy": {
      "command": "skyboy-mcp",
      "type": "stdio"
    }
  }
}

Node-at-runtime

This package shells out to npx, so Node.js must be on your PATH. It does not ship a pure-Python MCP implementation. That is deliberate: the brief specifies one MCP core (TypeScript) with a PyPI wrapper so Python users reach the same tools without a second implementation to maintain.

If you would rather avoid the wrapper, install the npm core directly:

npm install -g @skyboy/mcp-server

Tool surface (stdio, full mode)

• search_skills(query, category?, agent?) - fuzzy search slug/name/description/tags
• get_skill(slug) - SKILL.md + metadata + permissions
• get_plugin(slug) - plugin manifest + permissions (index + link only, never vendor-copied)
• list_categories() - taxonomy tree
• check_updates(installed_slugs[]) - compare installed versions to the catalog
• install_skill(slug, target_dir?) - write skill files to a local folder (stdio only)

The hosted remote endpoint (https://mcp.skyboy.in) exposes only the read-only tools (search, get, list, check_updates); install_skill is local-only because it writes to your filesystem.

License

MIT