experimental-ext-skills

This MCP server is part of the Skills Over MCP initiative and is focused on exploring how agent skills can be discovered, described, and consumed through MCP primitives. It demonstrates patterns for distributing skills as MCP-enabled assets, enabling host applications to discover available skills, activate them when appropriate, and coordinate multi-server workflows that involve tools and subagents. The server is designed to experiment with metadata, activation semantics, and patterns for skill orchestration across multiple MCP servers. You’ll typically run this alongside other MCP servers to test discovery, metadata exchange, and cross-server orchestration of agent skills. The available capabilities emphasize loading skills in a structured, machine-readable form, exposing metadata about each skill (descriptions, inputs, outputs, and dependencies), and enabling clients to select and activate skills as needed. In practice, you can use it to explore how host applications and MCP clients discover skills, fetch their definitions, and coordinate execution across multiple services.

Usage scenarios include discovering a set of skills provided by different servers, requesting a particular skill to be activated for a given agent, and coordinating tool invocations that span multiple servers as part of a larger workflow. The tooling focuses on metadata-driven discovery, experimental activation semantics, and patterns for progressive disclosure of skill capabilities to clients.

Prerequisites:\n- Python 3.9+ installed on your system.\n- Git installed if you’re cloning the repository.\n- Optional: pipx for isolated tooling.\n\nInstall from the package registry (recommended):\n1) Install the package via pipx to keep it isolated (if available):\n pipx install modelcontextprotocol-experimental-ext-skills\n # or the exact package name provided by the repo maintainers\n2) Run the module directly (assuming the installed package exposes the module):\n python -m modelcontextprotocol.experimental_ext_skills\n\nAlternative: install from source (if you prefer cloning the repo):\n1) git clone https://github.com/modelcontextprotocol/modelcontextprotocol-experimental-ext-skills.git\n2) cd modelcontextprotocol-experimental-ext-skills\n3) python -m modelcontextprotocol.experimental_ext_skills\n\nPrerequisites recap:\n- Python 3.9+\n- Access to install packages from PyPI or from the repository\n- Optional: virtual environment support (venv) to isolate dependencies

Tips and caveats:\n- This is marked as Experimental; expect changes in metadata formats and activation semantics as you experiment.\n- The server is designed for discovery and activation experimentation; production deployments should coordinate with Registry WG guidance.\n- If you plan to run multiple skill servers, consider aliasing or naming conventions to keep experiment scopes clear.\n- Ensure MCP_LOG_LEVEL is set to an appropriate level (e.g., INFO or DEBUG) when troubleshooting.\n- If you encounter import or module path issues, verify the module name matches the installed package structure (modelcontextprotocol.experimental_ext_skills).\n- For debugging, running with verbose logging and observing the MCP registry interactions can help validate discovery flows and activation paths.