Cloudways expands AI agent access with 244 hosting tools

Cloud hosting provider Cloudways has significantly expanded the capabilities of its AI agent integration, allowing automated systems to perform nearly all hosting account functions without manual intervention. The change reflects a broader industry shift toward delegating infrastructure management to AI-driven tools, though security researchers warn the approach carries substantial risks if not properly constrained.

What changed

Cloudways released version 1.2.112 of its Model Context Protocol (MCP) server this week, nearly doubling the number of available tools to 244. The MCP server acts as an intermediary between AI agents—such as Claude, Cursor, or Windsurf—and hosting accounts, enabling direct execution of tasks like service restarts, code deployments, security scans, and billing operations. Previously, these actions required manual input through the Cloudways dashboard.

The update introduces three permission tiers for API tokens: read-only for monitoring, limited scope for specific tasks, and full access for comprehensive control. This replaces the previous single-key system, which granted blanket permissions. Cloudways stated the change was intended to reduce the risk of overprivileged access, though the company did not specify when the new version was deployed.

Security concerns

The expansion of AI agent access comes amid growing scrutiny of MCP server vulnerabilities. Security firm BlueRock recently scanned over 10,000 public MCP servers and found that 9.2% contained critical flaws, while 43% were susceptible to command injection attacks. Such vulnerabilities could allow external actors to execute arbitrary commands on the server, potentially compromising broader infrastructure.

In one documented case, BlueRock researchers exploited a separate flaw in Microsoft’s MarkItDown MCP server to extract AWS access keys from a cloud instance’s metadata. Depending on the permissions associated with those keys, the breach could have provided access to the entire AWS account. While Cloudways’ new permission tiers may mitigate some risks, the underlying architecture remains exposed to similar attack vectors if not properly secured.

Industry context

Cloudways is not the first provider to implement AI-driven hosting management. Its parent company, DigitalOcean, introduced its own MCP server in August 2025, and community-developed versions exist for other control panels, including cPanel. The cPanel integration currently exposes 164 tools, demonstrating the rapid proliferation of this technology across the hosting ecosystem.

The primary concern for operators is not the number of tools available but the potential for unintended consequences when AI agents are granted broad access. A single compromised agent could become a single point of failure for multiple client sites, particularly for agencies managing numerous accounts. The shift toward scoped permissions reflects an industry-wide effort to balance automation with security, though the effectiveness of these measures remains unproven at scale.

What to watch

Hosting providers and agencies adopting AI agent integrations should prioritize the following:

• Token scoping: Ensure API tokens are restricted to the minimum required permissions for each task.
• Server hardening: MCP servers should be treated as high-risk components, with regular vulnerability scans and access controls.
• Audit trails: Maintain detailed logs of AI agent actions to detect and investigate unauthorized activity.

The long-term impact of AI-driven hosting management will depend on whether security practices evolve as quickly as the technology itself. For now, operators must weigh the convenience of automation against the risks of exposing critical infrastructure to new attack surfaces.