Products & Toolsgooglegeminigemini apiai developer tools

Google Expands Managed Agents in Gemini API

|July 9, 2026|By LDS Team

6.7

Relevance Score

Google Expands Managed Agents in Gemini API — Photo: Google Blog · rights & takedowns

Google said on July 7, 2026 that Managed Agents in Gemini API now support background execution, remote MCP servers, custom function calling, and credential refresh across interactions. For developers, the practical shift is that long-running agent jobs can keep working after the original client connection ends, reconnect to private tools through MCP, and preserve sandbox state while credentials are refreshed. That reduces custom orchestration around agent runtimes, especially for coding, research, and enterprise automation systems where tasks span many steps and external services. The update is not a new frontier model, but it is meaningful platform plumbing for teams moving agent prototypes into reliable workflows.

Google's Managed Agents update matters because it turns Gemini agent work from a request-response demo into a more durable runtime pattern. The useful shift is operational: background jobs, remote tool access, local function handoffs, and credential refresh reduce the custom queueing, proxy, and state-management code teams otherwise build around agent prototypes.

What happened

Google said on July 7, 2026 that Managed Agents in Gemini API now support background execution, remote MCP server integration, custom function calling, and credential refresh across interactions. With background execution, an application can start a long-running interaction and receive an ID for polling, streaming progress, or reconnecting later. Remote MCP support lets the managed agent connect to external tools and private systems from the sandbox, while custom functions hand work back to local business logic when an interaction requires action.

Technical context

The feature set points at agent runtime reliability, not model quality. Google describes managed agents as configurable agents that can use code execution, Google Search, URL context, mcp_server, and custom function tools. Its docs also say credential refresh can replace network rules for a new interaction while preserving the base environment's sources, repositories, installed packages, and file state. That makes the environment model more useful for long tasks, but the same docs still frame managed agents as preview infrastructure with limits such as no versioning and no subagent nesting.

For practitioners

Treat this as production plumbing for narrower, instrumented agents rather than a reason to hand broad autonomy to a model. Teams still need explicit policy around which MCP servers an agent can reach, how short-lived credentials rotate, when a local function can execute, and what logs prove a background job did the intended work. The value is less glue code around agent loops; the risk is putting private tools and credentials behind weak approval boundaries.

What to watch

The next signal is whether Google hardens observability, rollback, and governance around the Interactions API as managed agents move beyond preview. Durable background work is useful only if teams can audit the steps, cancel or resume safely, and constrain tool access without building a separate control plane.

Key Points

1Google added background execution, remote MCP integration, custom functions, and credential refresh to Managed Agents in Gemini API.
2The update matters for production agents because jobs can keep running, reconnect, and retain sandbox state.
3Teams still need governance around MCP tools, credentials, local functions, and preview API changes safely.

Scoring Rationale

Official Google developer platform update that addresses practical agent-runtime gaps: background jobs, remote MCP tool access, local function handoffs, and credential refresh. It is not a frontier model release, but it is notable production infrastructure for Gemini-based agents and enterprise developer tooling.

MoreGoogle AI news