@skillbrickai/mcp-server

MCP (Model Context Protocol) server for discovering and installing skills from the SkillBrick AI platform. This lets AI agents like Claude search, browse, and install skills directly.

Available Tools

| Tool | Description | |------|-------------| | search_skills | Search and browse skills by query, domain, tag, or sort order. Auto-infers domain and technology from natural language queries. | | get_skill | Get full skill details including prompt content | | install_skill | Install a skill (requires auth token) | | upload_skill | Upload or update a skill (upsert, requires auth token) | | sync_skills | Bulk-sync multiple skills in one call (requires auth token) | | my_skills | List all skills owned by the authenticated user | | check_credits | Check download credit balance | | list_collections | Browse skill collections | | get_collection | Get a collection with all its skills | | list_domains | List available skill domains/categories | | recommend_skills | Intelligent skill recommendations based on a task description |

Setup

Build

pnpm install
pnpm --filter @skillbrickai/mcp-server build

Claude Desktop

Add to your Claude Desktop config file (claude_desktop_config.json):

{
  "mcpServers": {
    "skillbrickai": {
      "command": "node",
      "args": ["C:/Users/user/Documents/skillbrickai/packages/mcp-server/dist/index.js"],
      "env": {
        "SKILLBRICK_API_URL": "https://skillbrickai.com",
        "SKILLBRICK_API_TOKEN": "your-api-token-here"
      }
    }
  }
}

Claude Code

claude mcp add skillbrick -- node C:/Users/user/Documents/skillbrickai/packages/mcp-server/dist/index.js

Or via npx (once published):

claude mcp add skillbrick -- npx @skillbrickai/mcp-server

Environment Variables

| Variable | Description | Default | |----------|-------------|---------| | SKILLBRICK_API_URL | Base URL for the SkillBrick API | https://skillbrickai.com | | SKILLBRICK_API_TOKEN | API auth token (required for install_skill, upload_skill, sync_skills, my_skills, check_credits) | (none) |

How the search works

1. Intent parsing -- extracts domain and technology tags from natural language
2. Domain inference -- maps casual terms ("backend", "k8s") to actual domain values
3. Multi-pass search -- queries by text, inferred domain, and inferred tags in parallel
4. Relevance scoring -- scores results by name/description/tag match + popularity signals
5. Fallback broadening -- if filtered search returns nothing, retries without filters