OpenClaw is an open registry of agent instruction files — system prompts, skills, workflows, domain packs, safety filters, and orchestration patterns. This MCP server wraps the registry's HTTP endpoints and exposes them as structured tools.

Tools:

search_registry — search by keyword, category, quality score
get_instruction — fetch a full instruction file (or just the deployable text)
list_categories — see all categories with file counts
get_featured — get all quality 90+ featured files

Install

npm install -g agentbase-mcp

Or use npx (no install):

npx agentbase-mcp

Configure

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "openclaw": {
      "command": "npx",
      "args": ["agentbase-mcp"]
    }
  }
}

Claude Code

claude mcp add openclaw npx agentbase-mcp

Continue (VS Code)

{
  "experimental": {
    "modelContextProtocolServers": [
      {
        "transport": {
          "type": "stdio",
          "command": "npx",
          "args": ["agentbase-mcp"]
        }
      }
    ]
  }
}

Any MCP Host

{
  "command": "npx",
  "args": ["agentbase-mcp"]
}

Usage Examples

Once connected, you can ask your AI agent:

"Search OpenClaw for safety filter instructions"
"Find a customer support system prompt"
"What categories of instruction files are available?"
"Get the structured web research skill and show me just the instruction text"
"Find featured orchestration patterns"

Direct API (no MCP)

The registry also has HTTP endpoints:

GET https://openclaw-sandy-eight.vercel.app/registry/index.json
GET https://openclaw-sandy-eight.vercel.app/registry/[slug].md
GET https://openclaw-sandy-eight.vercel.app/api/search?q=research&min_quality=85
GET https://openclaw-sandy-eight.vercel.app/llms.txt