skillport

SkillPort is an MCP-enabled server and CLI that helps you validate, manage, and deliver agent skills at scale. It supports a search-first loading pattern for a large catalog of skills, loading full instructions for a skill only when requested. This server exposes an MCP interface so MCP-compatible clients can query and load skills on demand, while also offering a CLI workflow to validate skills, add from various sources, and manage metadata. You can run SkillPort in MCP mode to serve skills to clients like Cursor, VS Code Copilot, Codex, and other MCP-enabled agents. The MCP surface is configured to point to a local skills directory, enabling per-project or shared SkillPort instances depending on your environment. To integrate, your MCP client should reference the skillport server via the uvx-based command and the specified environment variable for the skills location. The CLI on the host machine remains available for local management tasks such as validation, adding, updating, and inspecting skills.

Prerequisites:

• Python 3.10+ (per README standards)
• pip (Python package installer)
• uv (or uvx tool for MCP setup)

1. Install SkillPort (CLI + MCP tooling) using Python:

pip install skillport

2. Install the MCP server package (for MCP mode) if you want the dedicated MCP server tooling:

uv tool install skillport-mcp

3. Verify installation:

skillport --version
uv tool list

4. Start/prepare MCP configuration (example shown in README):

• The MCP server entry should be defined as:
  • command: uvx
  • args: ["skillport-mcp"]
  • env: { "SKILLPORT_SKILLS_DIR": "~/.skillport/skills" }

5. If you prefer running directly in a container or different environment, ensure the skills directory exists and contains your SkillPort skills:

mkdir -p ~/.skillport/skills
# populate with your skills or link to your existing directory

Notes and tips:

• SKILLPORT_SKILLS_DIR defines where SkillPort loads and serves skills. Update as needed for your environment.
• SKILLPORT_ENABLED_CATEGORIES, SKILLPORT_ENABLED_SKILLS, and SKILLPORT_ENABLED_NAMESPACES can be used to filter which skills are exposed via MCP in different client scenarios.
• When using MCP with multiple clients or projects, you can create separate mcpServers entries (e.g., skillport-dev, skillport-prod) each with its own env and filters.
• If you encounter issues with loading skills, verify that the SKILLPORT_SKILLS_DIR path is accessible by the process and that permissions allow reading the skill metadata and instructions.
• The SkillPort features include Validate, Manage, Metadata, and Serve, with Serve implementing a search_skills(query) and load_skill(skill_id) pattern to minimize upfront context load.
• For CLI usage, you can validate all skills with skillport validate or validate a specific directory with skillport validate ./skills.