SkillSync MCP
Uses Cloudflare AI for semantic search across the SkillsMP marketplace to find relevant skills based on natural language queries.
Enables scanning and installing skills from GitHub repositories, with security checks and automatic download to the skills directory.
Works with GitHub Copilot as an MCP-compatible client to search, scan, install, and manage skills via the SkillsMP marketplace.
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
SkillSync MCP
An MCP (Model Context Protocol) server for SkillsMP -- the marketplace for Claude Code skills. Search, scan for security threats, install, and manage skills directly from your AI assistant.
Every skill is scanned against 101 regex threat patterns before it installs; critical matches are blocked. The scanner is rule-based (no machine-learning model ships in this package): it can miss novel or obfuscated attacks and can flag benign text.
Research (experimental, not shipped)
Experimental detector: 99.2% attack detection on Boundary Pairs synthetic test data, in each of three training runs (119/120 attacks each). Average benign false-alarm rate 3.9% (per-run 2.5%, 4.2%, 5.0%). 10 held-out scenarios from one synthetic generator; not evidence of real-world protection. Shipped rule-based scanner is separate.
The detector is a fine-tuned DeBERTa-v3-base classifier. It is not part of the npm package and is not used by the install gate. Methods, results, baselines and negative results: docs/research/ and skillsync.js.org/research.html.
Related MCP server: skillsmp-mcp-server
Features
Search -- Keyword and semantic search across the SkillsMP marketplace (via the third-party SkillsMP API)
Security Scan -- 101 regex threat patterns in 19 categories: prompt injection, reverse shells, credential theft, supply chain attacks, crypto mining, obfuscation, and more
Install -- Download skills from GitHub to
~/.claude/skills/with automatic security gateUninstall -- Clean removal of installed skills
Safe Search -- Combined search + auto-scan in one step
Installed Skills Registry -- List all installed skills with risk levels and content hashes
Deep Audit -- Force a fresh security scan on any installed skill
Startup Verification -- Background discovery, content hashing, and
fs.watchfor live sync
Why SkillSync?
Raw | SkillSync MCP | |
Security scan before install | No | Regex scan -- 101 patterns |
Blocks critical findings | No | Install blocked on a critical pattern match (e.g. prompt injection, RCE, credential theft) |
Multi-client support | N/A | Config docs for Claude Code, OpenClaw, Cursor, Windsurf, GitHub Copilot, Zed, nanobot |
Marketplace search | Manual | Keyword + semantic search via the SkillsMP API |
Startup verification | No | fs.watch + content hash |
Output sanitization | No | Strips hidden Unicode controls from third-party text |
Data Flow
Search, suggestions and subscription sync send queries (and SKILLSMP_API_KEY, if set) to the SkillsMP API at skillsmp.com. Note that skillsmp_suggest builds its query from the names of up to 10 of your installed skills, plus the optional project context you pass (up to 200 characters), and sends that query to SkillsMP. Semantic search is provided by SkillsMP (it states it uses Cloudflare AI; the model is not disclosed). Skill source is fetched from github.com for scanning. Sync subscriptions live in a local config file; SkillSync has no cloud service of its own.
Tools (13)
Tool | Description |
| Keyword search across SkillsMP marketplace |
| Semantic search via the SkillsMP ai-search API (third-party) |
| Security scan a GitHub skill repo (101 regex patterns) |
| Search + auto-scan top results |
| Scan then install to |
| Remove an installed skill |
| List all installed skills with risk levels (optional refresh) |
| Deep security audit of a specific installed skill |
| AI-powered skill recommendations based on installed skills |
| Side-by-side security comparison of two skills |
| Manage sync subscriptions and settings |
| Run sync cycle: poll, diff, install/update/remove |
| Show sync engine status and schedule |
Compatible With
Works with Claude Code | OpenClaw | Cursor | Windsurf | GitHub Copilot | Zed | nanobot -- any MCP-compatible client
Install
Claude Code
Add to ~/.claude/settings.json:
{
"mcpServers": {
"skillsmp": {
"command": "npx",
"args": ["-y", "@stranzwersweb2/skillsync-mcp"]
}
}
}OpenClaw
Add to ~/.openclaw/mcp.json:
{
"mcpServers": {
"skillsmp": {
"version": "1.4.1",
"autoUpdate": false,
"command": "npx",
"args": ["-y", "@stranzwersweb2/skillsync-mcp@1.4.1"]
}
}
}OpenClaw uses the same
SKILL.mdformat as Claude Code. Skills installed via this server are compatible with both platforms. OpenClaw users should pin versions and review tool policies per the security hardening guide.
Cursor
Add to .cursor/mcp.json in your project root:
{
"mcpServers": {
"skillsmp": {
"command": "npx",
"args": ["-y", "@stranzwersweb2/skillsync-mcp"]
}
}
}Windsurf
Add to ~/.windsurf/mcp.json:
{
"mcpServers": {
"skillsmp": {
"command": "npx",
"args": ["-y", "@stranzwersweb2/skillsync-mcp"]
}
}
}GitHub Copilot
Add to .github/copilot-mcp.json in your project root, or ~/.github/copilot-mcp.json for global config:
{
"mcpServers": {
"skillsync": {
"command": "npx",
"args": ["-y", "@stranzwersweb2/skillsync-mcp"],
"env": {
"SKILLSMP_API_KEY": "your-api-key"
}
}
}
}Zed
Add to ~/.config/zed/settings.json under the "context_servers" key:
{
"context_servers": {
"skillsync": {
"command": {
"path": "npx",
"args": ["-y", "@stranzwersweb2/skillsync-mcp"],
"env": {
"SKILLSMP_API_KEY": "your-api-key"
}
}
}
}
}Global install
npm install -g @stranzwersweb2/skillsync-mcpThen reference in any MCP config:
{
"mcpServers": {
"skillsmp": {
"command": "skillsync-mcp"
}
}
}Client Compatibility
Client | Config Path | Skill Format |
Claude Code |
|
|
OpenClaw |
|
|
Cursor |
| MCP tools only |
Windsurf |
| MCP tools only |
GitHub Copilot |
| MCP tools only |
Zed |
| MCP tools only |
nanobot | MCP config | MCP tools only |
Security Model
Installation is gated by a rule-based, multi-level security scan: 101 regex patterns (47 critical, 3 critical multi-line, 51 warning) in 19 categories. See docs/THREAT_PATTERNS.md.
Risk Level | Behavior |
Safe / Low | Install proceeds, warnings shown |
Medium / High | Install blocked -- requires |
Critical | Install blocked -- no override |
Additional Safety Guards
Path traversal prevention on skill names and filenames
SSRF prevention -- only
github.comURLs acceptednpm install --ignore-scripts-- blockspostinstallattacksMax 50 files, 2MB total size limit
Binary files skipped, suspicious filenames flagged
Content hash for TOCTOU verification
Output sanitization -- strips zero-width Unicode, bidi overrides, truncates to prevent prompt injection
How It Works
Search SkillsMP -> Pick a skill -> Security scan (101 regex patterns)
|
Critical? -> BLOCKED
Medium/High? -> Requires force=true
Safe/Low? -> Download from GitHub
|
Write to ~/.claude/skills/<name>/
|
npm install --ignore-scripts (if needed)
|
Startup verification (fs.watch + content hash)
|
Restart your MCP client to loadExamples
Ask your AI assistant:
Search for git-related skills on SkillsMPScan this skill for security issues: https://github.com/user/repo/tree/main/skills/my-skillInstall the commit skill from https://github.com/user/repo/tree/main/skills/commitList all my installed skillsRun a deep security audit on the commit skillUninstall the commit skillDevelopment
git clone https://github.com/adityasugandhi/skillsync-mcp.git
cd skillsync-mcp
npm install
npm run build
npm run dev # Watch mode with tsx
npm run test:build # Build + run testsRequirements
Node.js >= 20
Any MCP-compatible client (Claude Code, OpenClaw, Cursor, Windsurf, GitHub Copilot, Zed, nanobot, etc.)
Contributing
Contributions are welcome. Please see CONTRIBUTING.md for guidelines.
Browse open issues or look for the
good-first-issuelabelTo add new threat detection patterns, see docs/THREAT_PATTERNS.md
All PRs must pass the existing test suite (
npm run test:build)
Publisher
The npm package is published under the legacy scope @stranzwersweb2 — this is the same author as the GitHub owner adityasugandhi. Both identities belong to Aditya Sugandhi.
Author
Aditya Sugandhi -- adityasugandhi.com | GitHub
Star History
License
This server cannot be deployed
Maintenance
Related MCP Connectors
Marketplace and infrastructure for MCP servers and skills: deploy, host, monetize, analytics.
MCP server for skill documentation, generated by doc2mcp.
Skills, agents and memory in the cloud, plus a gateway to your other MCP servers.
Search and install 4,000+ security-scanned MCP servers from inside any MCP-aware AI client.
Related MCP Servers
- AlicenseAqualityDmaintenanceAn MCP server for discovering Claude Code skills, plugins, and MCP servers by searching 15,000+ resources from 17 registries, GitHub, and the web with zero setup.344 npm3MIT
- AlicenseAqualityDmaintenanceThis MCP server enables AI agents to search, discover, and install skills from the SkillsMP marketplace, with support for keyword and semantic search, skill content retrieval, and installation to various coding agents.522 npm3MIT
- AlicenseNot gradedqualityDmaintenanceA powerful MCP server that brings the skills.sh ecosystem directly to your AI agents, enabling effortless discovery, installation, and management of skills.1MIT
- AlicenseAqualityDmaintenanceA lightweight MCP server that enables AI assistants to search and read skills from the SkillsMP marketplace with optional security scanning.343 npm1MIT