Skills-ContextManager

Skills-ContextManager exposes MCP tools to load and manage AI skills and workflows via a self-hosted web UI and an MCP-enabled server. It supports two loading modes: always_loaded skills that are preloaded into the agent's context, and dynamic skills that can be loaded on demand. The server provides three core tools for your AI agents: get_default_skills(), which loads all skills configured as always_loaded; list_available_skills(), which enumerates enabled skills along with their descriptions and loading modes; and load_full_skill_context(name), which loads a specific dynamic skill into the agent’s context when needed. These tools enable hub-aware context management across multiple workspaces and support both skills and workflows. To run with MCP, you typically start the server via the included setup script, then configure your agent’s MCP settings to reference the Skills-ContextManager entry and its Python execution path, as demonstrated in the provided MCP configuration example.

Prerequisites:

• Python 3.8+ installed on the host
• Basic Git and shell access
• Node.js (only if you plan to use the web UI locally; otherwise optional for backend MCP setup)

1. Clone the repository:

git clone https://github.com/One-Man-Company/Skills-ContextManager.git
cd Skills-ContextManager

2. Set up the Python environment and install dependencies:

python -m venv .venv
source .venv/bin/activate  # on Unix/macOS
# Windows: .venv\Scripts\activate
pip install -r requirements.txt

3. Run the setup to prepare MCP server script and configuration:

chmod +x setup.sh
./setup.sh

4. Start the web UI (if you want the local management interface):

cd Skill-ContextManager
npm install
npm start

Access the UI at http://localhost:3000

5. Run the MCP server (example using the provided mcp_server.py):

# From the repository root, ensure your virtual env is active, then run the MCP server
python -u mcp_server.py

6. Add the MCP configuration to your AI agent as shown in the MCP Configuration section of the README. The agent will connect using the python-based MCP server entry.

Tips and common considerations:

• Keep the virtual environment path accurate in your MCP config; use absolute paths if your agent requires them.
• The MCP tools (get_default_skills, list_available_skills, load_full_skill_context) assume the skills and workflows are correctly loaded into Skill-ContextManager and that the hub context is properly configured.
• If you enable dynamic loading, ensure network access to any external skill sources (Git, Skills.sh, etc.) is allowed by your host environment.
• For security, ensure that the agent MCP system prompt contains the MCP_instructions content as described, and that API keys or secrets are stored with restricted permissions (0600) as supported by the app.
• When upgrading, re-run the setup script to regenerate mcp_settings.json if paths or environments change.
• The web UI (Skill-ContextManager) is optional for MCP operation; MCP can run headless with the Python server if you prefer a script-based workflow.