Google Expands Gemini API Managed Agents with Background Execution and Remote MCP Support
Summary
- • Gemini API Managed Agents gain background async execution, ending reliance on fragile long-held HTTP connections for long-running tasks
- • Remote MCP server integration lets agents connect directly to private databases and internal APIs from their isolated sandbox
- • Custom function calling added alongside built-in sandbox tools with step-matching for server-side vs. client-side execution
- • Network credential refresh enables key rotation while preserving sandbox filesystem, installed packages, and repo state
Details
Background async execution
Pass background:true to run agent interactions asynchronously; API returns an ID immediately for polling, streaming, or reconnecting while the agent finishes on the server.
Remote MCP server integration
Agents connect directly to remote MCP servers—accessing private databases or internal APIs—without custom proxy middleware. MCP tools mix with built-in sandbox capabilities.
Custom function calling with step matching
Built-in sandbox tools run automatically server-side; custom functions trigger a requires_action state so the client executes local business logic, then resumes the interaction.
Network credential refresh
Pass existing environment_id with a new network config to rotate credentials or API keys; sandbox preserves filesystem state, installed packages, and cloned repos.
Isolated cloud sandbox environment
Managed Agents handle reasoning, code execution, package installation, file management, and web access inside an isolated cloud sandbox via a single API endpoint.
New capabilities announced for the Gemini Interactions API Managed Agents
What This Means
Google's Gemini Interactions API updates address the most common friction points developers face when building production AI agents—unreliable long-running connections, enterprise data access, and credential management. Remote MCP integration is particularly significant: it lets agents securely tap into private enterprise infrastructure without any custom middleware, dramatically reducing the engineering lift for enterprise adoption. Together these updates push the Gemini Managed Agents platform meaningfully closer to being a reliable, production-grade autonomous agent runtime, intensifying competition with other agentic platforms targeting enterprise developers.