# Gemini API adds tool retrieval preview to load function schemas on demand

> The Gemini Interactions API now searches large tool libraries and loads function schemas on demand instead of sending every tool on each request.

- Status: official (From the company, authors or authority themselves, with a link to their own page.)
- Topic: Products
- Published: 2026-11-19T00:00:00+00:00
- Updated: 2026-11-20T10:08:17.964754+00:00
- Web page: https://ai-news.ikangai.com/story/gemini-api-adds-tool-retrieval-preview-to-load-function-schemas-on-demand/
- Seen in: Gemini API changelog, Google AI
- Change: capability (New capability)
- Applies to: Gemini API, Interactions API, Gemini 3.6 and later, Gemini 4 models, remote MCP servers

## What changed

The Gemini API now has a preview of tool retrieval, so the Interactions API loads function schemas on demand instead of sending every tool with each request.

Google has added a preview of tool retrieval to the Gemini API. It lets the Interactions API search a developer's tool library and load function schemas only when a request needs them, rather than sending every tool with each call. The Gemini API changelog for November 19, 2026 describes the feature as a preview and is the only source for the details below.

The mechanism sits in the request itself. A developer marks each tool that should wait with defer_loading set to true, then adds a {"type": "tool_search"} tool. Gemini uses that search tool to retrieve the schemas it needs. The alternative is setting execution to "client", which hands retrieval to the developer's own logic. Remote MCP servers also accept defer_loading. As a result, large tool sets no longer have to be sent in full on every request.

The changelog limits the feature to a specific set of models. Tool retrieval works with Gemini 3.6 and later, including all Gemini 4 models, in the Interactions API. Requests to other models or endpoints are outside what the entry covers.

## Why it matters

Applications with large tool sets no longer need to send every function schema on every request, which can reduce per-request token use for agent builders.

## What you can do

If your agents carry many tools, test defer_loading with tool_search on a preview model before the feature leaves preview.

## How the story developed

- 2026-11-20: Adds that it works with Gemini 3.6 and later models, and that remote MCP servers support defer_loading.

## Original source

- [Introducing Tool Retrieval in the Gemini API](https://blog.google/innovation-and-ai/technology/developers-tools/gemini-api-tool-retrieval/)
- [Gemini API changelog: November 19, 2026](https://aistudio.google.com/docs/changelog)
- [Tool retrieval](https://aistudio.google.com/docs/tool-retrieval)

---

AI wrote this summary from the sources listed, and nobody reviewed it individually before publication. Check the original source before you rely on it.

Cite as: IKANGAI AI News, "Gemini API adds tool retrieval preview to load function schemas on demand", 2026-11-19, https://ai-news.ikangai.com/story/gemini-api-adds-tool-retrieval-preview-to-load-function-schemas-on-demand/
