SiteSkills MCP Server

MCP server that connects AI browser agents to the SiteSkills.ai marketplace. Search, fetch, and use pre-built navigation skills that teach agents how to navigate any website reliably.

Tools

| Tool | Description | |------|-------------| | siteskills_search | Search the marketplace by query, domain, category, agent compatibility | | siteskills_for_domain | Find the best skill for a specific domain + task (primary tool) | | siteskills_get | Get full skill details including SKILL.md navigation instructions | | siteskills_download | Get download URL for skill zip package | | siteskills_report_execution | Report execution success/failure (requires API key) |

Setup

Install

npm install
npm run build

Environment Variables

# Optional — enables telemetry reporting
export SITESKILLS_API_KEY=sk_live_xxx

# Transport: "stdio" (default) or "http"
export TRANSPORT=stdio

# Port for HTTP transport (default: 3000)
export PORT=3000

Claude Code

Add to your Claude Code MCP config (~/.claude/mcp.json):

{
  "mcpServers": {
    "siteskills": {
      "command": "node",
      "args": ["/path/to/siteskills-mcp-server/dist/index.js"],
      "env": {
        "SITESKILLS_API_KEY": "sk_live_xxx"
      }
    }
  }
}

Claude Desktop / Cowork

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "siteskills": {
      "command": "node",
      "args": ["/path/to/siteskills-mcp-server/dist/index.js"],
      "env": {
        "SITESKILLS_API_KEY": "sk_live_xxx"
      }
    }
  }
}

HTTP Mode (for remote / multi-client)

TRANSPORT=http PORT=3000 node dist/index.js

Then point any MCP client at http://localhost:3000/mcp.

Usage Flow

The typical agent workflow:

1. User says "add this to my Amazon cart"
2. Agent calls siteskills_for_domain with domain="amazon.com", task="add to cart"
3. Server returns the best skill with full SKILL.md navigation instructions
4. Agent follows the step-by-step instructions to navigate the site
5. Agent calls siteskills_report_execution with the result

License

MIT