{
  "id": 20783472,
  "title": "Introducing Tool Retrieval in the Gemini API",
  "url": "https://urgent.news/2026/11/19/introducing-tool-retrieval-in-the-gemini-api",
  "topic": "ai",
  "section": "AI",
  "published": "2026-11-19T18:00:00.000Z",
  "source": {
    "name": "Google AI Blog",
    "slug": "google-ai-blog",
    "url": "https://blog.google/innovation-and-ai/technology/developers-tools/gemini-api-tool-retrieval/"
  },
  "original_language": "en",
  "account": "Introducing Tool Retrieval in the Gemini API allows developers to manage extensive tool libraries more efficiently. By loading parameter schemas on demand, this feature reduces latency and input token costs for complex tasks. To enable Tool Retrieval, include a tool_search tool in your request and set defer_loading to true for any tools you wish to defer loading. The model retains knowledge of all deferred tools, enabling it to select the appropriate one when needed. When a tool is required, the API loads its full schema, and the model executes the function within the same request. This approach streamlines the tool selection process, as all three steps occur within a single API call. In practice, normal function calling can continue, with your code running the selected tool and sending back a function_result. No additional indexing is required on your part, as Gemini handles the search server-side. The only code changes needed are adding the new tool and setting the defer_loading flag for the desired tools. Deferred schemas remain invisible at the beginning of the prompt, ensuring consistent prefix caching throughout the conversation. The feature also works seamlessly with remote MCP servers, allowing you to load tools on demand without listing them individually. If you maintain your own tool registry or apply per-user permissions, you can define your search parameters using execution: client, enabling mid-conversation tool addition. Tool Retrieval operates under the same input and output token pricing as standard model calls, with a slight fixed token overhead for searches and billed input tokens for each loaded schema. The savings become apparent when dealing with large tool libraries, making Tool Retrieval an invaluable tool for developers working with the Gemini API.",
  "summary": "Give your agents access to hundreds of tools while using fewer tokens with Tool Retrieval, now in preview in the Gemini API.",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}