Gemini API managed agents add background work and external tools
Google added new features to in the Gemini API. A managed agent lets one API call handle reasoning, code execution, package installation, file handling, and web information inside an isolated workspace. lets long tasks keep running on the server while an app receives an ID, checks status, streams progress, or reconnects later.
Remote MCP server support lets agents connect directly to outside tools such as private databases or . Built-in sandbox tools and custom functions can now be used together; when a custom function is needed, the interaction pauses so the client app can run its own . Credentials such as short-lived access tokens or API keys can be refreshed in the next interaction by sending a new network , while the sandbox keeps its files, installed packages, and cloned repositories.
The examples use the @google/genai JavaScript SDK, with separate for Python and cURL.
Key points
- in the Gemini API now support , remote MCP servers, custom functions, and credential refresh.
- Long-running work can continue on the server while the app polls, streams progress, or reconnects later.
- Remote MCP servers let agents use outside tools such as private databases or .
- Custom functions pause the interaction so the client app can run local .
- Credential updates can replace old network rules without losing sandbox files, packages, or cloned repositories.