🪨 why use many token when few token do trick — Claude Code skill that cuts 65% of tokens by talking like caveman
CLI and MCP server for Agent Skills
SkillPort manages Agent Skills so agents can find, load, and use them without manual copying. It works in two modes: a CLI for shell-capable agents and an MCP server for clients that support tool connections.
Builders who want one skill library to work across Claude Code, Codex, Cursor, Copilot, Windsurf, and other MCP clients.
You can keep skills validated, organized, and available to different agents without redoing setup for each tool.
What it does
Validate skills
Checks skills against the Agent Skills specification, including missing fields, naming issues, and spec violations.
Manage skill lifecycle
Adds skills from GitHub, local paths, or zip files, then lists, updates, and removes them.
Edit metadata from the CLI
Lets you get, set, and unset skill metadata without opening the files.
Serve skills on demand
Exposes `search_skills(query)` and `load_skill(skill_id)` through MCP so clients can search first and load only what they need.
Generate project docs
Creates an `AGENTS.md` skill table for CLI-based agents to discover what is available.
Filter skills per client
Uses config and environment variables to expose different skills, categories, or namespaces to different agents.
How to get it
- 1Run
uv tool install skillport # or: pip install skillport
- 2Run
# Add from GitHub skillport add anthropics/skills skills # Or use a custom skills directory skillport --skills-dir .claude/skills add anthropics/skills skills
- 3Run
skillport validate # ✓ All 5 skill(s) pass validation
README
⚓ SkillPort
Why SkillPort?
| When you... | SkillPort... |
|---|---|
| Use a coding agent without native skill support | Serves via MCP or CLI |
| Build your own AI agent | Provides MCP server, CLI, and Python library |
| Have 50+ skills and need the right one fast | Search-first loading (Tool Search Tool pattern) |
| Check skills before deployment | Validates against the spec in CI |
| Manage skill metadata programmatically | Provides meta commands |
| Find a skill on GitHub | Installs with add <url> |
Fully compatible with the Agent Skills specification.
Features
Validate
Check skills against the Agent Skills specification.
skillport validate # Validate all skills
skillport validate ./skills # Validate specific directory
skillport validate --json # CI-friendly JSON output
Catches missing fields, naming issues, and spec violations before they cause problems.
Manage
Full lifecycle management from any source.
# Add from GitHub (shorthand)
skillport add anthropics/skills skills
# Add from GitHub (full URL)
skillport add https://github.com/anthropics/skills/tree/main/skills
# Add from local path or zip
skillport add ./my-skills
skillport add skills.zip
# Update, list, remove
skillport update # Update all from original sources
skillport list # See installed skills
skillport remove <skill-id> # Uninstall
Metadata
Update skill metadata without editing files manually. Useful for automation and keeping skills consistent across a team.
skillport meta get <skill> <key> # Get metadata value
skillport meta set <skill> <key> <val> # Set metadata value
skillport meta unset <skill> <key> # Remove metadata key
Serve
MCP server for clients that don't natively support Agent Skills.
Inspired by Anthropic's Tool Search Tool pattern — search first, load on demand:
| Tool | Purpose |
|---|---|
search_skills(query) | Find skills by description (full-text search) |
load_skill(skill_id) | Get full instructions + path |
Why search matters: With 50+ skills, loading all upfront consumes context and hurts accuracy. SkillPort loads metadata only (~100 tokens/skill), then full instructions on demand.
Works with Cursor, Copilot, Windsurf, Cline, Codex, and any MCP-compatible client.
Quick Start
Install
uv tool install skillport
# or: pip install skillport
Add Skills
# Add from GitHub
skillport add anthropics/skills skills
# Or use a custom skills directory
skillport --skills-dir .claude/skills add anthropics/skills skills
Validate
skillport validate
# ✓ All 5 skill(s) pass validation
Connect to Agents
Choose how to deliver skills to your AI agents:
| Mode | Best for | Setup |
|---|---|---|
| CLI Mode | Agents with shell access (Cursor, Windsurf, Codex, etc.) | Per-project |
| MCP Mode | MCP-compatible clients, multi-project | One-time |
CLI Mode
For agents that can run shell commands. No MCP configuration required.
skillport init # Initialize project
skillport doc # Generate AGENTS.md with skill table
skillport show <id> # Load full instructions for a skill
How it works:
skillport docgenerates a skill table in AGENTS.md- The agent reads AGENTS.md to discover available skills
- When needed, the agent runs
skillport show <id>to load full instructions
MCP Mode
For MCP-compatible clients. Install the server:
uv tool install skillport-mcp
Add to your client's config:
{
"mcpServers": {
"skillport": {
"command": "uvx",
"args": ["skillport-mcp"],
"env": { "SKILLPORT_SKILLS_DIR": "~/.skillport/skills" }
}
}
}
One-click install for popular clients
Cursor
VS Code / GitHub Copilot
Kiro
CLI Agents
# Codex
codex mcp add skillport -- uvx skillport-mcp
# Claude Code
claude mcp add skillport -- uvx skillport-mcp
Organization
Organize skills with categories and tags. Works with both CLI and MCP modes.
Categories & Tags
Use metadata to organize and filter skills:
# SKILL.md frontmatter
metadata:
skillport:
category: development
tags: [testing, quality]
alwaysApply: true # Always available (Core Skills)
Per-Client Filtering
Expose different skills to different agents:
{
"mcpServers": {
"skillport-dev": {
"command": "uvx",
"args": ["skillport-mcp"],
"env": { "SKILLPORT_ENABLED_CATEGORIES": "development,testing" }
},
"skillport-writing": {
"command": "uvx",
"args": ["skillport-mcp"],
"env": { "SKILLPORT_ENABLED_CATEGORIES": "writing,research" }
}
}
}
Configuration
| Variable | Description | Default |
|---|---|---|
SKILLPORT_SKILLS_DIR | Skills directory | ~/.skillport/skills |
SKILLPORT_ENABLED_CATEGORIES | Filter by category | all |
SKILLPORT_ENABLED_SKILLS | Filter by skill ID | all |
SKILLPORT_ENABLED_NAMESPACES | Filter by namespace | all |
SKILLPORT_CORE_SKILLS_MODE | Core Skills behavior (auto/explicit/none) | auto |
Creating Skills
Create a SKILL.md with YAML frontmatter:
---
name: my-skill
description: What this skill does
metadata:
skillport:
category: development
tags: [example]
---
# My Skill
Instructions for the AI agent.
See Creating Skills Guide for best practices.
Skill Sources
| Source | Features | Target | URL |
|---|---|---|---|
| Anthropic Official | Document skills (docx, pdf, pptx, xlsx), design, MCP builder | All users | GitHub |
| Awesome Claude Skills | Curated community collection, 2.5k+ stars | Discovery | GitHub |
| Hugging Face Skills | Dataset creation, model evaluation, LLM training, paper publishing | ML/AI engineers | GitHub |
| Claude Scientific Skills | 128+ scientific skills (bio, chem, ML), 26+ databases | Researchers | GitHub |
| ClaudeKit Skills | 30+ skills, auth, multimodal, problem-solving frameworks | Full-stack devs | GitHub |
| Superpowers | TDD, debugging, parallel agents, code review workflows | Quality-focused devs | GitHub |
| Kubernetes Operations | K8s deployment, monitoring, troubleshooting | DevOps/SRE | GitHub |
Learn More
Development
Status: Work in progress. APIs may change.
git clone https://github.com/gotalab/skillport.git
cd skillport
uv sync
# Run MCP server
SKILLPORT_SKILLS_DIR=.skills uv run skillport-mcp
# Run CLI
uv run skillport --help
# Run tests
uv run pytest
License
MIT
Files in the repo
- .github
- .skills
- guide
- packages
- src
- tests
- .env.example
- .gitignore
- .release-please-manifest.json
- .skillportrc
- AGENTS.md
- CHANGELOG.md
- CLAUDE.md
- LICENSE
- pyproject.toml
- README.md
- release-please-config.json
- uv.lock
- verify_server.py
Discussion (0)
Ask about usage, or say what you built with itSign in to join the discussion.
No comments yet. Be the first to say what this is good for.
More tools
The best-benchmarked open-source AI memory system. And it's free.
Orca is the ADE for working with a fleet of parallel agents. Run any coding agent with your own subscription. Available on desktop, mobile and remote runtime.

A cross-platform desktop All-in-One assistant for Claude Code, Codex, OpenCode, OpenClaw, Grok Build & Hermes Agent. Only official website: ccswitch.io
Never stop coding. Free MIT AI gateway: one endpoint, 352 providers (150+ free), 1200+ models Kimi, Claude, GPT, Gemini, GLM, DeepSeek, MiniMax. Works with Claude Code, Codex, Cursor, OpenCode, Cline & Copilot. Quota-aware auto-fallback, RTK+Caveman compression saves 15-95% tokens, MCP/A2A, Desktop/PWA. Built by 550+ contributors
Compress tool outputs, logs, files, and RAG chunks before they reach the LLM. 20% fewer tokens for coding agents, 60-95% fewer tokens for JSON, same answers. Library, proxy, MCP server.