News
November 20, 2026 · Google

Gemini API adds Tool Retrieval so agents load tool schemas only when needed

Google's new Tool Retrieval preview in the Gemini Interactions API lets agents keep large tool libraries on standby and load full schemas on demand, at no extra charge.

Google introduced Tool Retrieval for the Gemini API on November 19, 2026. According to a post by Gemini API product manager Lucia Loher and Google DeepMind research scientist Nora Kassner, the feature is in preview in the Interactions API and lets an agent hold its tools in reserve, fetching a tool's full parameter schema only when a task calls for it.

What's new

  • The problem it targets: Google says that when an agent is connected to dozens or hundreds of tools, sending every schema with every request raises input token counts, slows responses and makes the model worse at choosing the right tool.
  • How to turn it on: developers add a tool_search tool to the request and mark the tools they want held back with defer_loading: true. The model still sees each deferred tool's name and description; it requests a full schema by name, and the search, schema load and function call all happen inside one request, according to the post.
  • MCP support: Google says the same flag works on a remote MCP server entry, so every tool that server exposes is loaded on demand.
  • Client-side option: teams with their own tool registry or permission filters can run the search themselves by setting "execution": "client". Google notes this costs one extra round trip per search.
  • Caching: because deferred schemas are left out of the opening part of the prompt, Google says the cached prefix stays stable across turns.

Availability and pricing

Google says there is no extra fee for Tool Retrieval; tokens are billed at the normal rate of whichever model is used. Switching on the search tool brings a modest, constant input-token overhead, and any schema the model loads counts as input for the rest of the conversation. It works with Gemini 3.6 and later models, including the Gemini 4 family. Developers can try it through Google AI Studio.

Why it matters

Analysis: large tool libraries are a common pain point for agent builders, and similar on-demand loading is appearing across the industry. Google itself advises measuring savings against real traffic, since the benefit depends on how many schemas an agent never needs to load.

Source: Google