Google DeepMind has announced a significant expansion of Managed Agents in the Gemini API, adding four major capabilities that transform managed agents from simple chat interfaces into reliable, production-ready asynchronous workers.
The updates address real developer feedback and include:
Background Execution: Developers can now pass `background: true` to run long-running agent interactions asynchronously on the server. The API immediately returns an ID that client applications can use to poll for status, stream progress, or reconnect later — eliminating the need to hold HTTP connections open for extended tasks.
Remote MCP Server Integration: Instead of writing custom proxy middleware to access private databases or internal APIs, developers can now connect managed agents directly to remote Model Context Protocol (MCP) servers. This allows agents to communicate with enterprise endpoints from within their secure sandbox environment.
Custom Function Calling: Teams can add custom tools alongside built-in sandbox tools for local execution. The API uses step matching — built-in tools run automatically on the server, while custom functions transition the interaction to `requires_action` so client code can execute local business logic.
Network Credential Refresh: Access tokens and short-lived API keys can now be refreshed or rotated by passing an existing `environment_id` with a new network configuration on subsequent interactions. The sandbox preserves its filesystem state, installed packages, and cloned repositories during credential rotation.
These updates build on the Gemini Interactions API, where a single endpoint handles reasoning, code execution, package installation, file management, and web information inside an isolated cloud sandbox. The announcement includes examples using the `@google/genai` JavaScript SDK, with Python and cURL documentation available in the Antigravity agent docs.
The expansion positions Google's managed agents more competitively against Anthropic's Claude Code and OpenAI's Codex as enterprise-grade agentic platforms.