mcp-link-server

The project emphasizes strong security, privacy, and user control. It features explicit warnings about agentic AI risks (data loss, unauthorized actions) in the EULA, a 'Zero-Collection' privacy policy, and detailed logging/auditing of all tool actions. It supports sandboxing for critical operations (e.g., Docker integration for the `docker` tool). Sensitive operations like `execute_command` (via `system` tool) and arbitrary Python code execution (via `python` tool) are explicitly provided as powerful features, but are designed to be used with granular permissions, user approval prompts, and sandboxing. Dynamic generation of `TOOL_UNLOCK_TOKEN` for each tool enhances security by preventing static hardcoded tokens and ensuring context awareness. Input validation against schemas is consistently applied across tools. The centralized contribution model and 'visible source' philosophy (despite proprietary license) indicate a strong focus on code integrity and security audits.