Can See

Can See is an open-source MCP server that enables AI agents to see and interact with terminal/CLI applications. By launching apps in virtual terminals, capturing PNG screenshots of the terminal, and sending keystrokes or text input, it empowers agents—like Claude Code or any MCP-compatible agent—to visually debug, automate, and manage interactive terminal applications. It's especially useful for scenarios where textual output alone isn’t sufficient, such as debugging TUI apps or automating CLI wizards where the visual terminal state matters.