0xharryriddle/core-actionbook

Actionbook

Pre-computed action manuals for browser automation. Agents receive structured page information instead of parsing entire HTML.

Workflow

1. search_actions - Search by keyword, returns URL-based action IDs with content previews
2. get_action_by_id - Get full action manual with page details, DOM structure, and element selectors
3. Execute - Use returned selectors with your browser automation tool

MCP Tools

• search_actions - Search by keyword. Returns: URL-based action IDs, content previews, relevance scores
• get_action_by_id - Get full action details. Returns: action content, page element selectors (CSS/XPath), element types, allowed methods (click, type, extract), document metadata

Parameters

search_actions:

• query (required): Search keyword (e.g., "airbnb search", "google login")
• type: vector | fulltext | hybrid (default)
• limit: Max results (default: 5)
• sourceIds: Filter by source IDs (comma-separated)
• minScore: Minimum relevance score (0-1)

get_action_by_id:

• id (required): URL-based action ID (e.g., example.com/page)

Example Response

{
  "title": "Airbnb Search",
  "url": "www.airbnb.com/search",
  "elements": [
    {
      "name": "location_input",
      "selector": "input[data-testid='structured-search-input-field-query']",
      "type": "textbox",
      "methods": ["type", "fill"]
    }
  ]
}