source: google ai: expanding managed agents in gemini api: background tasks, remote mcp and more
level: technical
google deepmind announced new capabilities for managed agents in the gemini interactions api. developers can now run agents asynchronously in the background, connect directly to remote model context protocol servers, mix custom function calling with built-in sandbox tools, and refresh network credentials without losing sandbox state. these updates respond to developer feedback and aim to make agents more reliable for production use.
the background execution feature lets clients pass a flag to run interactions asynchronously. the api returns an id immediately, and clients can poll for status, stream progress, or reconnect later. this avoids holding open http connections for long-running tasks. remote mcp server integration removes the need for custom proxy middleware. developers can pass an mcp_server tool at interaction time alongside built-in tools like google search or code execution, letting the agent access private databases or internal apis from its secure sandbox.
custom function calling works through step matching: built-in tools run automatically on the server, while custom functions pause the interaction and require client-side execution. credential refresh allows passing an existing environment id with new network rules on the next interaction, replacing old credentials immediately while keeping the sandbox filesystem, installed packages, and cloned repositories intact. these changes turn managed agents into asynchronous workers that operate inside real development environments without blocking applications.
why it matters: these updates let developers build more autonomous and reliable ai agents that can handle long tasks, access private data securely, and maintain state across sessions, reducing infrastructure complexity.
source: google ai: expanding managed agents in gemini api: background tasks, remote mcp and more