skill-fetch

mcp
Security Audit
Pass
Health Pass
  • License — License: MIT
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Community trust — 23 GitHub stars
Code Pass
  • Code scan — Scanned 12 files during light audit, no dangerous patterns found
Permissions Pass
  • Permissions — No dangerous permissions requested
Purpose
This MCP server acts as a search engine and package manager for AI coding agents. It allows users to search, score, and install custom "skills" across nine different registries directly into tools like Claude Code, Cursor, and others.

Security Assessment
Overall Risk: Medium. The tool is designed to discover and execute external code, which inherently carries risk. To its credit, it includes strong built-in mitigations: it performs a 6-category security scan on downloaded files, actively checks for prompt injections, and uses SHA-256 hash verification to detect file tampering. It also relies on shell scripts and avoids exposing API keys in process arguments. No dangerous patterns, hardcoded secrets, or excessive permissions were found in the tool's own code. However, the risk remains moderate because its core function involves fetching and executing third-party scripts from the internet.

Quality Assessment
The project is actively maintained, with its most recent code push happening today. It is fully licensed under the permissive and standard MIT license. It has garnered 23 GitHub stars, indicating a small but growing level of community trust and real-world usage. A recent CI workflow is also in place, showing good automated testing practices.

Verdict
Use with caution—the tool itself is well-structured and secure, but always review the third-party skills and scripts you choose to install through it.
SUMMARY

Multi-registry skill discovery and installation for AI coding agents — search 9 sources, score, paginate, and install agent skills with security labels

README.md

skill-fetch

License: MIT
GitHub release
GitHub stars
CI

Cross-platform skill discovery and installation for AI coding agents

Search, score, and install AI agent skills from 9 registries in parallel — works across Claude Code, Cursor, Codex, Gemini CLI, Windsurf, and Amp.

Quick Start

  1. Install — pick any method from Installation below
  2. Set up API keys — run /fetch-skill-config (interactive, recommended) or configure manually
  3. Search & install skills/fetch-skill react native animation

Features

  • 9 Search Sources — SkillsMP (semantic + keyword), GitHub, Anthropic Skills, ClawSkillHub, skills.sh, PolySkill, SkillHub, Skills Directory
  • Cross-Platform — Works on 6+ AI coding agents with automatic tool adaptation
  • Multi-Variant AI Search — 3 query variants fired in parallel, improving recall significantly
  • Quality Scoring — 0-100 composite score: Relevance (40) + Freshness (25) + Community (20) + Trust (15) + External Bonus (5)
  • Security Labels — 5 trust tiers: Official, Verified, Partial, Unverified, Security Concerns
  • 6-Category Security Scan — Destructive commands, RCE, data exfiltration, system modification, obfuscation, prompt injection
  • Prompt Injection Detection — 5 sub-categories (PI-1~PI-5): direct override, hidden role markers, encoding tricks, indirect injection, social engineering
  • Complete Bundle Install — Downloads the whole skill directory (SKILL.md + references/, scripts/, assets/, templates/, prompts/, data/, etc.), preserving the subtree layout so skills that delegate to sibling files actually work
  • Integrity Verification — SHA-256 hash recorded for every file in the bundle at install; tamper detection on future loads
  • Paginated Results — Browse 5 at a time with c to continue; install by number from any page
  • Local/Global Install — Choose project-level or user-level installation (mandatory prompt)
  • Deduplication — Same skill across registries is merged; similar descriptions are flagged
  • Bundled Scripts — Shell scripts for SkillHub and Skills Directory APIs (no API key exposure in argv) plus fetch-skill-bundle.sh for full-bundle GitHub installs

Installation

Choose the method that fits your setup:

Option 1: Plugin (Best experience, Claude Code)

# Add the skill-fetch marketplace
claude plugin marketplace add girofu/skill-fetch

# Install skill-fetch plugin
claude plugin install skill-fetch@skill-fetch

Provides auto-updates, /fetch-skill command, and full plugin integration.

Option 2: npx skills add (Node.js)

npx skills add girofu/skill-fetch

Works with Claude Code, Cursor, and Codex.

Option 3: curl | sh (Universal, zero dependencies)

curl -fsSL https://raw.githubusercontent.com/girofu/skill-fetch/main/install.sh | bash

Auto-detects installed agents and installs for all of them. Specify a single agent:

curl -fsSL https://raw.githubusercontent.com/girofu/skill-fetch/main/install.sh | bash -s -- --agent claude

Option 4: Python installer

python3 -c "$(curl -fsSL https://raw.githubusercontent.com/girofu/skill-fetch/main/install.py)"

Or download and run:

curl -fsSL https://raw.githubusercontent.com/girofu/skill-fetch/main/install.py -o install.py
python3 install.py --agent cursor

Option 5: Manual (git clone)

git clone https://github.com/girofu/skill-fetch.git
# Claude Code
cp -r skill-fetch/skills/skill-fetch ~/.claude/skills/skill-fetch

# Cursor
cp -r skill-fetch/skills/skill-fetch ~/.cursor/skills/skill-fetch

# Other agents: replace ~/.cursor/ with ~/.codex/, ~/.gemini/, ~/.windsurf/, or ~/.amp/

⚙️ Setup API Keys

Important: API keys unlock the full 9-source search. Without them, only Sources 3-8 are available (still useful, but fewer results).

Interactive Setup (Recommended)

Run /fetch-skill-config in Claude Code. It will:

  • Show which keys are currently configured
  • Walk you through setting each key with links to get them
  • Automatically install the SkillsMP MCP server if needed
  • Write everything to ~/.claude/skills/.fetch-config.json

Manual Setup

🔑 SkillsMP API Key (Sources 1-2) — Recommended

SkillsMP provides the best semantic search with AI-powered matching. Get your free API key:

  1. Go to skillsmp.com and create an account
  2. Copy your API key
  3. Register the MCP server:
claude mcp add --scope user skillsmp -- npx -y skillsmp-mcp-server --env SKILLSMP_API_KEY=your_key_here
  1. Restart Claude Code — SkillsMP tools will be available in the next session

🔑 SkillHub & Skills Directory (Sources 8-9) — Optional

For additional coverage, create ~/.claude/skills/.fetch-config.json:

{
  "SKILLSMP_API_KEY": "your-skillsmp-key",
  "SKILLHUB_API_KEY": "sk-sh-your-key-here",
  "SKILLS_DIRECTORY_API_KEY": "sk_live_your-key-here"
}

Sources 3-8 work without any API keys. Source 9 (Skills Directory) requires a key.


Supported Agents

Agent Plugin npx curl/sh Python Manual
Claude Code
Cursor
Codex
Gemini CLI
Windsurf
Amp

Usage

# Search for skills (Claude Code)
/fetch-skill react native animation

# Install from URL
/fetch-skill https://github.com/owner/repo

# Auto-triggered by skill-eval hook
# (no manual invocation needed)

In other agents, the skill activates automatically when referenced in context.

How It Works

┌─────────────────────────────────────────────────────┐
│                   User Query                         │
│              "testing React Native"                  │
└──────────────────────┬──────────────────────────────┘
                       │
        ┌──────────────┼──────────────────┐
        ▼              ▼                  ▼
   ┌─────────┐   ┌──────────┐      ┌──────────┐
   │SkillsMP │   │SkillsMP  │      │SkillsMP  │
   │AI Var A  │   │AI Var B  │      │AI Var C  │
   └────┬─────┘   └────┬─────┘      └────┬─────┘
        │              │                  │
        └──────┬───────┘──────────┬───────┘
               ▼                  ▼
   ┌───────────────┐    ┌─────────────────┐
   │ SkillsMP KW   │    │    GitHub        │
   │ + Anthropic    │    │    + PolySkill   │
   └───────┬───────┘    └────────┬────────┘
           │                     │
   ┌───────┴───────┐    ┌───────┴────────┐
   │  ClawSkillHub │    │  skills.sh     │
   └───────┬───────┘    └───────┬────────┘
           │                     │
   ┌───────┴───────┐    ┌───────┴────────┐
   │  SkillHub     │    │ Skills         │
   │  (API script) │    │ Directory      │
   └───────┬───────┘    └───────┬────────┘
           │                     │
           └──────────┬──────────┘
                      ▼
              ┌───────────────┐
              │  Deduplicate  │
              │  Score & Rank │
              │  (0-100)      │
              └───────┬───────┘
                      ▼
              ┌───────────────┐
              │  Security     │
              │  Scan (A-F)   │
              └───────┬───────┘
                      ▼
              ┌───────────────┐
              │  Display 5    │
              │  per page     │
              └───────┬───────┘
                      ▼
              ┌───────────────┐
              │  User picks   │
              │  → Install    │
              │  → Hash       │
              └───────────────┘

Cross-platform note: SkillsMP sources require the MCP server + API key (see API Key Setup). On other platforms, the flow starts from GitHub + supplementary sources.

Search Sources

# Source Method Type Availability
1 SkillsMP (semantic) skillsmp_ai_search MCP × 3 variants Primary Claude Code + MCP
2 SkillsMP (keyword) skillsmp_search MCP Primary Claude Code + MCP
3 GitHub gh search code --filename SKILL.md (primary) + gh search repos (supplementary) Primary All agents
4 Anthropic Skills gh search code in anthropics/skills Official All agents
5 ClawSkillHub npx -y clawhub search Supplementary Agents with npx
6 skills.sh HTTP API / WebFetch Supplementary All agents
7 PolySkill npx -y @polyskill/cli search (single keyword) Supplementary Agents with npx
8 SkillHub Bundled shell script (API) or CLI fallback Supplementary All agents
9 Skills Directory Bundled shell script (API) Supplementary Configured agents

All sources are searched in parallel. Sources fail gracefully — if any is unavailable, the search continues with remaining sources.

API Key Setup

See ⚙️ Setup API Keys above for detailed instructions.

Quality Scoring

Each result receives a composite score (0-100):

Dimension Weight How
Relevance 0-40 LLM judges description-to-task semantic match
Freshness 0-25 Time since last GitHub push
Community 0-20 Star count (log scale)
Trust 0-15 Source credibility tier
External Bonus 0-5 Security/quality signals from PolySkill, SkillHub, Skills Directory

Grade labels:

Score Label
85+ 🟢 Strongly Recommended
70-84 🟢 Recommended
55-69 🟡 Worth Considering
40-54 🟡 Marginal
<40 🔴 Not Recommended

Security

Security Labels

Label Criteria
🔒 Official From anthropics/skills repo
🔒 Verified SkillsMP stars ≥ 50 + securityGrade A/B + scan clean
⚠️ Partial Lower stars but scan clean, or standard frontmatter present
⚠️ Unverified Direct URL or no external signals
⚠️ Security Concerns Scan found issues or securityGrade D/F

6-Category Security Scan

All skills are scanned before (or immediately after) installation:

Category Name Severity
A Destructive Commands Critical
B Remote Code Execution Critical
C Data Exfiltration High
D System Modification High
E Obfuscation Medium
F Prompt Injection (5 sub-types) High

Integrity Verification

After installation, SHA-256 hashes are recorded for all skill files. On future loads, hashes are compared to detect tampering.

Permissions Declaration (Advisory)

Skills may optionally declare their required permissions (network, filesystem-write, shell-commands, external-urls) in SKILL.md frontmatter. The scanner flags mismatches between declared and actual behavior.

File Structure

skill-fetch/
├── .claude-plugin/
│   └── plugin.json                   # Claude Code plugin manifest
├── commands/
│   ├── fetch-skill.md               # /fetch-skill slash command
│   └── fetch-skill-config.md        # /fetch-skill-config API key setup
├── skills/
│   └── skill-fetch/
│       ├── SKILL.md                  # Main skill (cross-platform)
│       ├── references/
│       │   ├── search-sources.md     # Source-specific commands & dedup rules
│       │   ├── quality-signals.md    # Scoring algorithm + security labels
│       │   ├── interaction-patterns.md # Output templates, security scan, prompt injection
│       │   ├── platform-adapters.md  # Cross-platform tool mapping
│       │   ├── installation-guide.md # Complete Step 3 install workflow
│       │   └── local-index.md        # Local skill/plugin scan for pre-search dedup
│       └── scripts/
│           ├── fetch-skill-bundle.sh    # Download a complete skill bundle from GitHub (SKILL.md + all siblings)
│           ├── fetch-skillhub.sh        # SkillHub API search (reads key from config)
│           └── fetch-skills-directory.sh # Skills Directory API search
├── .github/
│   └── maintainer/                   # Per-repo state for open-source-maintainer workflow
├── install.sh                        # Universal bash installer
├── install.py                        # Python installer
├── CHANGELOG.md                      # Version history
├── README.md                         # This file
└── LICENSE                           # MIT

Requirements

Minimum (any agent):

  • Shell with curl or WebFetch for HTTP-based searches

Full experience (Claude Code):

  • SkillsMP MCP server + API key from skillsmp.com (see API Key Setup)
  • GitHub CLI (gh) for GitHub and Anthropic Skills search
  • Node.js for npx-based searches (ClawSkillHub, PolySkill, SkillHub CLI)
  • ~/.claude/skills/.fetch-config.json for SkillHub and Skills Directory APIs (optional)

License

MIT

Reviews (0)

No results found