AgentSkills

You can run AgentSkills MCP to extend any MCP-compatible agent with Anthropic’s Agent Skills. It loads skills on demand to fit within your context window, works across multiple transports, and lets you point to your own skill directory for quick setup.

How to use

You run AgentSkills MCP as a local process or expose it remotely, then connect your MCP client to load and use skills on demand. Start with a local stdio setup to integrate directly with your agent, or run an SSE/HTTP server to connect from remote clients. When you trigger a skill, the system loads the Skill’s content as needed, minimizing overhead until you actually use a capability.

How to install

Prerequisites: Python and a Python environment, plus an MCP client that can talk to MCP servers. Install the AgentSkills MCP package with Python’s package manager.

pip install mcp-agentskills

# Optional: for development if you want to modify the code
uv pip install mcp-agentskills

Additional content

Load Skills by creating a directory to store Skill packs, then populate it with Skill folders. Each Skill folder contains a SKILL.md file describing the Skill and how to use it.

Run modes include local process communication via stdio and a remote SSE/HTTP server. The following configurations show how to start and connect in each mode.

Environment variables required for running the local stdio mode include your API key and the base URL for the LLM service.

Security and notes

Keep your API keys secure. Use separate skill directories for development and production, and restrict access to the server endpoints. When running remotely, ensure proper network security and authentication as appropriate for your deployment.

Troubleshooting

If skills fail to load, verify that the Skill directory exists and contains valid SKILL.md files. Check that environment variables are set correctly and that the server transport (stdio or http) is reachable from your MCP client.

Available tools

load_skill_metadata_op

Loads the names and descriptions of all Skills into the agent context at startup (always invoked)

load_skill_op

Loads the content of a specific Skill when triggered by the agent

read_reference_file_op

Reads files from a skill, such as scripts or reference documents, on demand

run_shell_command_op

Executes shell commands to run executable scripts included in a skill, on demand