Wuying AgentBay Execution Infrastructure

English | 中文

Abstract

Alibaba Cloud's Wuying AgentBay presents a foundational cloud substrate tailored for the deployment of intelligent Agents. It furnishes organizations, independent developers, and AI solution providers with a turnkey provisioning utility and operational sandbox for complex Agent initiatives. Via the AgentBay MCP Gateway or the Wuying API structure, Agent constructs can seamlessly embed and leverage requisite ancillary functions, operating with inherent elasticity on the cloud fabric.

Context: What is Wuying?

Wuying (meaning 'Shadowless'), concurrently known as Alibaba Cloud Workspace, represents Alibaba Cloud's pioneering converged cloud-network computational paradigm. It delivers secure, high-performance digital workspaces via virtual machines, application streaming, and lightweight endpoint access. Core attributes include:

Robust data protection mechanisms
Negligible infrastructure upkeep overhead
Minimalist endpoint asset footprint management
Simplified interconnection capabilities
Globally distributed access with optimized latency characteristics

AgentBay Defined

AgentBay functions as the specialized cloud middleware underpinning Wuying for AI Agent deployment. It delivers:

Expedited, single-click provisioning tools for Agent task execution frameworks
A secured operational containerization layer
Enterprise-grade foundational services for vendors and developers
Accelerated integration pathways via Software Development Kits (SDKs) and the MCP Server interface
Native serverless execution capabilities

Principal Capabilities

Standardized Execution Context
Pre-configured essential utilities for Agent task fulfillment
Adherence to the Model Context Protocol (MCP) specification
Accelerated integration capacity for corporate IT environments
User Session State Retention
Secure partitioning and persistence of user preferences and authentication tokens (cookies)
On-demand mapping of preserved user contexts
Cloud execution environment closely mirroring local desktop configurations
Low-Latency Edge-to-Cloud Synchronization
Proprietary ASP streaming protocol engineered for real-time data transit
Awareness and managed redirection of local peripheral hardware
Dedicated channels for network and input/output stream redirection

Technical Blueprint

Stateful Persistence Layer: Custom non-volatile memory system engineered to retain operational states and artifact histories for users.
Communication Fabric: Proprietary ASP protocol ensuring synchronous, low-latency data exchange between the remote compute cluster and user terminals.
Scalability: Architecture capable of instantaneously provisioning up to 10,000 concurrent operational instances, leveraging Alibaba Cloud's extensive resource pools.
Deployment Model: Inherently serverless operation with streamlined, one-click lifecycle management for isolated session environments.
Utility Interoperability: Support for essential MCP-compliant tools: Web Navigator, File System Manager, and Command Shell.
Integration Vectors: Compatible with both direct SDK/API invocations and the standardized MCP Server communication method.

Operational Service Parameters

Computational Assets

The platform allows utilization of shared compute resources. A limited complimentary tier is available during the public preview phase, capped at 10 simultaneous active sessions. Geolocation of service deployment is dynamically mapped based on the customer's access IP address.

Runtime Image Specifications

The platform natively supports encapsulated Linux distribution images. Wuying maintains an active pipeline for enhancing and updating these images while strictly preserving backward compatibility with older configurations.

Supported MCP Components

Initial deployment furnishes built-in support for the Browser, File, and Terminal utilities via the MCP framework. For the most current roster of supported tools, please consult the official service portal.

Initial Setup Guide

Phase 1: Credential Generation

Acquire your requisite authentication credentials (API Key).

Access the AgentBay Administration Portal.
Navigate to the Service Administration section in the left-hand navigation panel.
On the Service Administration view, initiate the creation process by clicking Generate New API KEY.
In the configuration modal, assign a descriptive label to the key and confirm by clicking OK.

Restriction: During the introductory phase, the maximum allowance is set to 10 unique API Keys per account. The newly issued key will populate the management list, from which it can be securely copied for integration.

Phase 2: MCP Framework Configuration

Verify the runtime environment selection and retrieve the necessary MCP connection parameters.

Log into the AgentBay Administration Portal.
Select Configure Operational Resources from the navigation menu.
Choose a previously generated credential from the API Key selection dropdown.
Select the desired operational image from the Select Image dropdown list.

Note: The service relies on pre-built Linux images. Wuying is committed to continuous enhancement of these images while guaranteeing adherence to prior version specifications.

Under the Tool Capabilities and Interface section, review the array of accessible MCP tools. Click Details View for comprehensive specifications.
Locate the Acquire MCP Information text area and utilize the Copy Snippet button in the upper-right corner. Integrate this configuration payload into any application supporting the MCP standard (e.g., Cline, Cursor).

An illustrative setup using Cursor Version 0.48.9 follows:

Access the Cursor Settings pane and navigate to the MCP tab on the left.
In the MCP Gateways panel, select Add New Global Gateway at the top.
Insert the subsequent configuration JSON into the opened mcp.json file and save. Supported connection protocols include SSE and STDIO.

Crucial: Replace the placeholder YOUR_API_KEY with the actual key starting with akm-, generated in Step 1.

Example employing the SSE streaming method: The SSE methodology establishes a persistent network link upon each service invocation, precluding the requirement for local plugin downloads.

{
  "mcpServers": {
    "wuying_mcp_server": {
      "url": "https://agentbay.wuying.aliyun.com/sse?APIKEY=YOUR_API_KEY"
    }
  }
}

Upon task completion, the cloud environment may be decommissioned. To mandate the preservation of configuration data or the operational state for subsequent use, append the EXTERNALID parameter to the connection string below:

{
  "mcpServers": {
    "wuying_mcp_server": {
      "url": "https://agentbay.wuying.aliyun.com/sse?APIKEY=YOUR_API_KEY&EXTERNALID=user001"
    }
  }
}

Phase 3: Screen Streaming Integration (Optional)

Remote desktop viewing is accessible either by embedding the Wuying Web SDK within an iframe or by direct client access via a standard browser. When invoking the cloud session stream, the MCP interface will return a secure, time-sensitive link structured as: https://wuying.aliyun.com?mcp.html?authcode=<authCode>&resourceId=<resourceId>. - This access token is transient and has defined usage limits; immediate utilization is advised. - Input control (mouse/keyboard) can be enabled via URL parameters such as &input=true&keyboard=true, or dynamically managed through the Web SDK methods. Comprehensive details are available in the Web SDK Reference Documentation。

Next Operational Steps

You can validate the setup by providing a directive such as launch the web browser using wuying-agentbay and navigate to wuying.aliyun.com. Upon successful configuration, the Wuying cloud instance will materialize, and browser actions will be executed remotely. During this active process, the underlying model will output a Web link allowing you to open and assume direct manual control over the Wuying session stream. Next Steps