Google Launches Tool Retrieval in Gemini API to Reduce Token Usage for Large Tool Libraries

·2026.11.20 03:00

Key point

The feature is available in preview for Gemini 3.6 and later models, allowing agents to defer loading parameter schemas until needed.

Details

  • Tool Retrieval is now in preview in the Gemini Interactions API, enabling agents to keep large tool libraries on standby and load parameter schemas only when a specific task requires them.
  • This approach reduces input token counts and latency by preventing the full tool library from being sent in every prompt, while maintaining tool-selection accuracy as the number of available tools scales.
  • Developers enable the feature by adding a tool_search tool and setting defer_loading: true on specific tools; the model sees tool names and descriptions but excludes schemas from the context until requested.
  • The feature supports both server-side retrieval (no index required) and client-side retrieval (for custom registries) and works with remote MCP servers by deferring all tools exposed by a server.
  • Pricing follows standard input and output token rates, with a small fixed block of input tokens added for the tool_search tool and loaded schemas billed as input for the remainder of the conversation.
  • Tool Retrieval is compatible with Gemini 3.6 and later models, including all Gemini 4 models.

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.