MCP-server-client-computer-use-ai-sdk

The system grants AI direct control over the operating system, allowing it to perform actions like launching applications, typing, clicking, and pressing keys. This is inherently a high-risk operation. While the client-side components (Next.js and CLI) communicate with the Rust-based desktop control server over a local HTTP connection (127.0.0.1:8080), a compromised AI model or malicious prompt could execute arbitrary, potentially destructive commands on the user's machine (e.g., 'delete all files'). The macOS server explicitly requires user-granted accessibility permissions, which is a necessary safeguard. No hardcoded API keys are found within the client code that would be exposed to the browser, as they are loaded from environment variables and used server-side in Next.js API routes. The core risk is the nature of giving AI full system control and the trust placed in the AI's reasoning and the input prompts.