Google previews Tool Retrieval in the Gemini API
Reported by Mezha and Google AI2 sources ↓
Google says Tool Retrieval is available in preview in the Gemini Interactions API. It loads tool schemas on demand, rather than including every schema in each request.
Key points
- It is available in the Interactions API with Gemini 3.6 and later, including all Gemini 4 models
- Developers can defer schemas and have the model request a tool’s full schema when needed
- Google says there is no separate charge; standard model token rates apply
Sources
Details
Google says developers can add a tool_search tool and mark selected tools for deferred loading. The model still sees each deferred tool’s name and description; when it requests one, the API loads its full parameter schema and the model can call it within the same request.
Google says the server handles retrieval without requiring developers to build or host an index. The feature also supports remote MCP servers. Developers using their own registry or search can instead run retrieval on the client, which Google says adds a round trip per search.
Google says tool_search adds a small fixed amount of input tokens, and each loaded schema is billed as input for the rest of the conversation. Token savings depend on how many schemas are not loaded, so the company recommends measuring usage against an application’s traffic.
Written from coverage as of
Images
- Google AI
- Google AI
Stocks in this story
Change since open: 20 Nov to 21 Nov 2026
Last refresh