Manages custom Agent resources on Gemini Enterprise Agent Platform. Use when the user wants to programmatically create, configure, list, update, or delete stateful, server-managed Agent resources (including mounting files, skills, and tools) before executing conversations.
Detected risks:
Instructions for programmatically managing custom Agent resources on the Gemini Enterprise Agent Platform, which forms the platform control plane. It is intended for creating, configuring, listing, updating, and deleting stateful, server-managed agent containers, including mounting files, skills, and tools, before any conversation is executed. The material provides REST endpoints and JSON payload structures rather than a wrapper abstraction.
All requests authenticate with a Bearer token from Application Default Credentials, obtained via gcloud auth print-access-token, and target the production global endpoint at aiplatform.googleapis.com v1beta1 scoped by project and location, where the location must be a region where the Managed Agents API is supported, such as global. Creating an agent is a POST that returns a Long-Running Operation; the payload sets an id, a base agent, a description, a system instruction, a tools array, and a base environment that can mount remote sources from Google Cloud Storage or, alternatively, from the Skill Registry by using a skill_registry source type, along with a network allowlist. Third-party Model Context Protocol servers can be attached by adding an entry with type mcp, a name, a url, and optional headers to the tools array, and the platform routes tool requests to that server while keeping the supplied headers confidential to that URL.
Because creation is asynchronous, the create call returns an operation tracking object whose name is polled with a GET against the operations endpoint until done is true and the completed Agent resource appears under response. Additional control-plane operations include getting an agent by id to retrieve its configuration, tools, and environment setup. The documentation also notes that MCP tools can be overridden at interaction time on the data plane by passing a type mcp_server entry when creating a conversation interaction, deferring those details to the Interactions API.
It manages custom Agent resources on the Gemini Enterprise Agent Platform control plane, letting you create, configure, retrieve, update, and delete stateful agent containers with mounted files, skills, and tools.
Every request includes a Bearer token from Application Default Credentials obtained with gcloud auth print-access-token, targeting the production global aiplatform.googleapis.com v1beta1 endpoint.
Creation is a Long-Running Operation that returns an operation name; you poll that operation with a GET until done is true, at which point the finished Agent resource appears under the response field.
Yes. Instead of a gcs source you can use a skill_registry source pointing at a Skill Registry revision path, mounted to a target such as the agent skills directory.
Add a tools entry with type mcp, a name, a url, and optional auth headers; the platform routes tool execution to that server and only sends the supplied headers to that URL.
Quick Setup:
.claude/skills/Repository
google/skills