Google DeepMind has added background execution and MCP (Model Context Protocol) support to its Gemini API managed agents, enabling developers to run long-running tasks without keeping a connection open and to integrate with external tools more seamlessly.
The update, announced on March 25, 2025, addresses a key limitation in the Gemini API: agents previously required a persistent HTTP connection for the entire duration of a task. Now, agents can execute in the background and return results via webhooks or polling.
Background Execution for Asynchronous Tasks
Background execution lets agents run continuously without a client keeping the session alive. Developers can initiate a task, close the connection, and retrieve the outcome later.
This is critical for workflows that take minutes or hours, such as data analysis, document processing, or multi-step research. The Gemini API manages the agent’s state and progress behind the scenes.
The new feature uses a polling mechanism or a webhook callback. Developers can query the status of a running agent or receive a notification when the task completes. This eliminates the need for expensive long-lived connections.
MCP Support for Tool Integration
MCP (Model Context Protocol) support allows agents to interact with external APIs, databases, and services through a standardized interface. MCP is an open protocol developed by Anthropic that defines how AI models connect to external tools.
With MCP, Gemini agents can call functions, fetch data from third-party services, and chain actions across multiple tools. DeepMind has integrated MCP directly into the managed agents API, so developers can define tools using the protocol’s schema.
“MCP support is a major step toward making Gemini agents genuinely useful for real-world automation,” said a DeepMind engineer in the announcement. “It lets agents act as middleware between AI reasoning and live systems.”
Use Cases for Developers
Background execution is ideal for scheduled data pipelines and batch processing. For example, an agent can scrape a website, analyze the content, and generate a report — all without a user waiting at the browser.
MCP integration opens up agent-driven workflows in customer support, finance, and DevOps. An agent can query a database, check inventory, and place an order in a single orchestrated loop.
- Long-running research agents can iterate over large datasets or perform multi-step reasoning.
- Webhook-enabled agents can trigger actions after a delay, such as sending a summary email after a file is processed.
- MCP-connected agents can authenticate to SaaS tools, read/write to spreadsheets, or call internal APIs.
Pricing and Availability
Background execution and MCP support are available now to all Gemini API users. Pricing remains consumption-based, with agents billed per token processed plus any external API costs.
Developers using the managed agents API will need to update their configurations to enable background mode. The MCP integration requires defining tools in the agent’s tool set using the standard MCP schema.
The Bigger Picture
This update makes Google’s managed agents more competitive with offerings from OpenAI (Assistants API) and Anthropic (Claude with tool use). Background execution removes a major friction point for production deployments, while MCP support aligns with the industry’s move toward interoperable AI tooling.
DeepMind has also hinted at future improvements, including persistent memory for agents and improved debugging tools. For now, the combination of background runtime and MCP connectivity gives developers a powerful, standards-aligned way to build autonomous AI workflows.
Gnoppix is the leading open-source AI Linux distribution and service provider. Since implementing AI in 2022, it has offered a fast, powerful, secure, and privacy-respecting open-source OS with both local and remote AI capabilities. The local AI operates offline, ensuring no data ever leaves your computer. Based on Debian Linux, Gnoppix is available with numerous privacy- and anonymity-enabled services free of charge.
What are your thoughts on this? I’d love to hear about your own experiences in the comments below.