home / mcp / skillport mcp server

SkillPort MCP Server

Provides an MCP interface to discover, load, and execute agent skills on demand.

Installation

Add the following to your MCP client configuration file.

Configuration

{
  "mcpServers": {
    "gotalab-skillport": {
      "command": "uvx",
      "args": [
        "skillport-mcp"
      ],
      "env": {
        "SKILLPORT_SKILLS_DIR": "~/.skillport/skills"
      }
    }
  }
}

SkillPort provides an MCP server interface that lets MCP-compatible clients load and deliver agent skills on demand. It uses a search-first loading approach to keep context lean and maintain high accuracy, while still offering a full CLI for managing skills locally and an HTTP-capable server path for MCP clients.

How to use

You connect to SkillPort through an MCP client to fetch skills on demand. The server exposes a lightweight load strategy: it keeps only essential metadata in memory and loads full instructions for a skill only when requested by the client. This minimizes context usage while preserving responsiveness and accuracy. You can run SkillPort in stdio (local) mode for a single client or via an HTTP-based MCP endpoint for multiple clients.

How to install

Prerequisites: make sure you have Node.js and/or Python tooling installed on your system depending on how you plan to run SkillPort. The MCP server is provided as a stdio configuration that you can run from your environment.

uv tool install skillport-mcp

# Run the MCP server (stdio/stdio-like runtime)
uvx skillport-mcp

Configuration and management

Configure SkillPort for your MCP clients by defining an MCP server entry that tells the client how to start SkillPort and where to find its skills. The following example shows a stdio configuration you can embed in your client's MCP config.

{
  "mcpServers": {
    "skillport": {
      "command": "uvx",
      "args": ["skillport-mcp"],
      "env": { "SKILLPORT_SKILLS_DIR": "~/.skillport/skills" }
    }
  }
}

Run and test the MCP server

Start the server via your MCP client configuration and verify it is reachable. Use the client’s mechanisms to query the available skills and then load a specific skill by its identifier when your agent needs it.

Additional usage notes

SkillPort supports both local CLI interactions and MCP client integrations. You can manage skills, validate them against the Agent Skills spec, and load them on-demand for your agents. If you want to customize skill delivery for multiple agents, you can create per-client MCP server configurations that enable different skill sets or namespaces.

Persistence and environment

SkillPort uses a local skills directory by default. You can override this with environment variables in your MCP client config to point to a different location where skills are stored or downloaded.

Where to find and read more

If you need to manage skills, you can extend SkillPort with its CLI subcommands for validation, adding, updating, listing, removing, and metadata operations. These utilities help you keep your skill set clean, consistent, and ready for deployment.

Available tools

validateSkills

Validate skills against the Agent Skills specification to ensure compliance before deployment.

addSkills

Add skills from a source (GitHub, local path, or zip) to the SkillPort catalog.

updateSkills

Update all skills from their original sources.

listSkills

List installed or available skills in your SkillPort environment.

removeSkill

Uninstall a specific skill by its identifier.

metaGet

Retrieve metadata values for a given skill key.

metaSet

Set metadata values for a given skill key to automate updates.

metaUnset

Remove a metadata key from a given skill.

loadSkill

Load full instructions for a skill on demand via the MCP interface.

searchSkills

Search skills by description or metadata to find relevant capabilities.