Google previews Tool Retrieval for Gemini API agents

Google has launched Tool Retrieval in preview for the Gemini API, a feature designed to help AI agents that connect to dozens or hundreds of tools work faster and cost less. The company announced the feature on 19 November in a blog post by Lucia Loher, a product manager on the Gemini API, and Nora Kassner, a research scientist at Google DeepMind.
Every tool an agent can call comes with a parameter schema describing its inputs, and until now those schemas were usually sent with every request. Google says that inflates input token counts, adds latency and makes the model worse at choosing the right tool as the list grows.
Schemas loaded only when needed
With Tool Retrieval, developers add a tool_search tool and mark tools with defer_loading: true. The model still sees each deferred tool's name and description, but the full parameter schema stays out of context until the model asks for it by name. The API then loads the schema and the model makes the call, all inside one request.
Google's example is a refund agent: a customer reports order A-1042 arrived damaged and asks for a 40-dollar refund. The model searches for the issue_refund tool, the schema loads, and the function call is made. Each step is recorded in the interaction record, which Google says makes it easier to debug why a tool was chosen or skipped.
There is no separate charge, and no index to build or host — Gemini runs the search on its servers. The feature works with Gemini 3.6 and later models, including the whole Gemini 4 family, but only in the Interactions API. It also works with remote MCP servers, and a client-side option lets teams run the search in their own app.