ProductGoogleConfirmed
Story 3 of 16 in the Thursday 19 November editionGoogle previews Tool Retrieval in Gemini Interactions API
Live on 20 November 2026 at no extra fee for Gemini 3.6 and Gemini 4 models, deferring function and remote MCP parameter schemas until needed.

Google launched Tool Retrieval in preview on 20 November 2026 on the v1beta Gemini Interactions API across Gemini 3.6 and all Gemini 4 models, including Gemini 4 Neon and Gemini 4 Hydrogen. Instead of passing tens of thousands of JSON Schema tokens on every request, developers declare a tool_search entry and set defer_loading: true (plus an optional short_description) on long-tail function declarations or remote mcp_server connections.
Server-side single-call schema resolution versus client-side registries
In default server-side mode, Gemini sees only the name and summary of deferred tools; when a task requires one, the model emits a tool_search_call step, the API injects the full schema in a tool_search_result step and Gemini emits the function_call inside the same HTTP round trip. Because deferred schemas stay out of the opening prompt, cached prompt prefixes remain intact across turns.
Applications with proprietary vector indexes or per-user tool permissions can set execution to client, adding one round trip to return dynamic function declarations mid-conversation. Google charges no separate fee for Tool Retrieval beyond standard model input and output token rates; generateContent is not supported.
Key details
- API surface and version
- Gemini Interactions API (v1beta preview); not supported on generateContent
- Supported Gemini models
- Gemini 3.6 and later models, including all Gemini 4 models
- Retrieval modes and MCP support
- Server-side (default, 0 extra round trips, remote MCP supported) and client-side
- Additional feature fee
- None; billed at standard model input/output rates for tool_search and loaded schemas
Release details & context
Stateless signature circulation and multi-turn declaration rules
Even when chaining turns via previous_interaction_id, developers must resend the complete tools array (including tool_search and every deferred function or mcp_server) on every request so Gemini retains visibility over uncalled tools. In stateless mode (store set to false), applications must replay every tool_search_call and tool_search_result step in its original position with its encrypted signature field untouched, or the follow-up turn fails with HTTP 400.
Google advises leaving frequently used tools and any function explicitly named inside system instructions undeferred so the model never invokes a deferred tool without searching first.
Common questions
Does Gemini API Tool Retrieval work with remote MCP servers and prefix caching?
Yes, setting defer_loading to true on an mcp_server entry defers every exposed tool on demand, and loaded schemas append later in the conversation so initial cached prefixes stay unchanged.
Sources
- PrimaryGoogle Blog: Introducing Tool Retrieval in the Gemini APIblog.google
- OfficialGoogle AI Studio Docs: Tool retrieval in the Gemini APIaistudio.google.com
- On XGoogle AI Studio (@GoogleAIStudio) Tool Retrieval announcement on Xx.com
Google release timeline (40 launches tracked)
- Google AI Studio adds custom domains and HD Android emulator builds
Google AI Studio adds direct custom-domain purchasing and hosting alongside HD audio Android emulation, Firestore sync and personal Play publishing.
- Google launches Gemini 4 Neon at $3 and $15 per million tokens
Google launches Gemini 4 Neon between Hydrogen and Carbon, ranking first on the Artificial Analysis Coding Agent Index at 58.6.
- Google Labs unveils Stitch Loop 24/7 product optimisation agent
Google Labs introduces Stitch Loop, an evergreen post-launch agent that monitors live product metrics and opens sandboxed GitHub pull requests.
- Google brings re-architected Gemini app to Android Go edition phones
Google launches a lower-memory build of the Gemini app for eligible Android (Go edition) phones with 45-language Gemini Live voice chat.

