Auto-discover and install the right skills from skills.sh before your AI agent builds anything.
An MCP server for Cursor, Claude Code, Windsurf, and any MCP-compatible agent. Tell it what you want to build — it finds the best skills, installs them to disk, and loads their knowledge into context automatically.
1. You type: "Build me a booking app with payments"
2. skill-pull searches skills.sh semantically
3. Installs matching skills to <project>/.cursor/skills/
4. Loads skill content into the agent's context
5. Agent builds with expert knowledge from skill authors
┌─────────────────────────────────────────────────────┐
│ AI Agent (Cursor etc.) │
│ │
│ User prompt ──► auto-skills rule triggers │
│ │ │
│ ┌────────▼────────┐ │
│ │ skill-pull │ ◄── MCP (stdio) │
│ │ MCP Server │ │
│ └────────┬────────┘ │
└───────────────────────┼─────────────────────────────┘
│
┌─────────────┼─────────────┐
▼ ▼ ▼
recommend_skills install_skill get_skill_details
│ │ │
▼ ▼ ▼
skills.sh disk write skills.sh
/api/search .cursor/skills /api/download
Agent receives user prompt
│
▼
[auto-skills.mdc rule fires]
│
▼
recommend_skills(query)
└─► GET skills.sh/api/search?q=<query>&limit=5
└─► Returns: [{id, name, description, installs}]
│
▼
install_skill(id, projectPath)
└─► Parse id → owner/repo + slug
└─► Try: npx skills add <source> --skill <slug>
└─► Fallback: GET skills.sh/api/download/<owner>/<repo>/<slug>
└─► Write files → <projectPath>/.cursor/skills/<slug>/
│
▼
get_skill_details(id)
└─► GET skills.sh/api/download/<owner>/<repo>/<slug>
└─► Extract SKILL.md content
└─► Return full text to agent context
│
▼
Agent builds using loaded skill knowledge
my-project/
└── .cursor/
└── skills/ ← travels with the repo
├── nextjs-best-practices/
│ └── SKILL.md
├── stripe-payments/
│ ├── SKILL.md
│ └── rules/
│ └── checkout.md
└── supabase-auth/
└── SKILL.md
skill-pull/
├── src/
│ ├── index.js # MCP server entry — wires tools + zod schemas
│ ├── skills-api.js # skills.sh API client (search + download)
│ └── tools/
│ ├── recommend.js # recommend_skills tool
│ ├── install.js # install_skill tool (npx + API fallback)
│ ├── get-details.js # get_skill_details tool
│ └── list-installed.js # list_installed_skills tool
├── agent-rules/ # Rule files for each agent type
│ ├── CLAUDE.md # Claude Code
│ ├── .windsurfrules # Windsurf
│ ├── AGENTS.md # OpenAI Codex / generic
│ └── copilot-instructions.md
├── setup.js # One-command rule installer for all agents
└── SKILL.md # skill-pull listed as a skill on skills.sh
| Decision | Reason |
|---|---|
| stdio transport | No hosting needed — Cursor spawns the process directly |
Project-level skills (<project>/.cursor/skills/) |
Skills travel with the repo; teammates & other agents pick them up automatically |
| npx fallback → direct API | Works even without the skills CLI installed |
id as primary input |
Agent passes the id from recommend_skills directly — no manual parsing |
| JavaScript, no build step | Runs anywhere Node 18+ is present; zero compile overhead |
Add to ~/.cursor/mcp.json:
{
"mcpServers": {
"skill-pull": {
"command": "npx",
"args": ["-y", "@rohangore1999/skill-pull"]
}
}
}Then run the setup script to install the auto-trigger rule for your agent:
npx @rohangore1999/skill-pull setupOr point to a local clone:
{
"mcpServers": {
"skill-pull": {
"command": "node",
"args": ["/path/to/skill-pull/src/index.js"]
}
}
}| Tool | Description |
|---|---|
recommend_skills |
Semantic search skills.sh for your build description |
install_skill |
Install skill into <projectPath>/.cursor/skills/ |
get_skill_details |
Load full SKILL.md content into agent context |
list_installed_skills |
Show all locally installed skills |
Once configured, just describe what you want to build. The agent handles the rest automatically:
"Build a SaaS app with auth and Stripe payments"
→ Finds: shadcn/ui, supabase, stripe-skill, next-best-practices
→ Installs each to <project>/.cursor/skills/
→ Loads expert knowledge into context
→ Builds with best practices from skill authors
Works with any MCP-compatible agent: Cursor, Claude Code, Claude Desktop, Windsurf, VS Code Copilot, Cline, Kiro, and more.
MIT