# Tool Retrieval — Gemini API agents load tool schemas only when needed

> Gemini agents can hold hundreds of tools and pay input tokens only for the ones they use.

- Org: Google
- Date: 2026-11-19
- Category: tool, ecosystem
- Source: https://blog.google/innovation-and-ai/technology/developers-tools/gemini-api-tool-retrieval/
- Canonical page: https://ai-tldr.dev/releases/google-gemini-api-tool-retrieval/

## In short

Google's Tool Retrieval lets Gemini API agents keep large tool libraries on standby and load a tool's full schema only when a task needs it. It is in preview in the Interactions API.

## What it is

Tool Retrieval is a Gemini API feature, in preview in the Interactions API, that defers loading of tool definitions. The model sees only each deferred tool's name and description, and asks for the full parameter schema when a task needs it.

## How it works

You add a tool_search tool and set defer_loading: true on the tools to hold back, including mcp_server tools. Gemini runs the search on the server, so no index is needed, and calls the function in the same request. Setting execution to client lets your own app run the search instead.

## Why it matters

Large tool libraries usually bloat the prompt, which costs tokens and can hurt tool choice. Deferred schemas stay out of the prompt prefix, so prefix caching keeps working across turns. Google gives no benchmark figures, so measure savings on your own traffic.

## Who it's for

developers building agents with many tools or MCP servers

## Try it

Add a tool_search tool and defer_loading: true to your tools in the Interactions API

## Links

- [Announcement](https://blog.google/innovation-and-ai/technology/developers-tools/gemini-api-tool-retrieval/)
- [Documentation](https://aistudio.google.com/docs/tool-retrieval)
- [Interactions API](https://aistudio.google.com/docs/interactions-overview)
- [Remote MCP Servers](https://ai.google.dev/gemini-api/docs/function-calling#mcp)
