Detailed Analysis
A user troubleshooting Claude Code has identified a specific configuration setting responsible for breaking terminal copy-and-paste functionality during SSH sessions. The culprit is the `"tui": "fullscreen"` option found in the `~/.claude/settings.json` configuration file. When enabled, this setting causes Claude Code to capture or alter mouse input behavior within the terminal, preventing users from selecting and copying text as they normally would with other command-line tools running in the same session. The user verified the fix by toggling the setting on and off multiple times across two different Linux distributions—Ubuntu and AlmaLinux—confirming a direct causal relationship between the fullscreen TUI mode and the broken clipboard behavior.
This issue, tracked as GitHub issue #62699 in the Anthropic/claude-code repository, highlights a common friction point in terminal-based developer tools: the tension between rich, immersive text-user-interface (TUI) experiences and standard terminal input handling. Fullscreen TUI modes typically take over the entire terminal viewport and often intercept mouse events to enable custom UI interactions like scrolling panels, clickable menus, or in-app selection. However, this interception can conflict with the terminal emulator's native mouse-reporting protocols, which are what allow users to highlight and copy text via standard OS-level clipboard mechanisms. Over SSH connections specifically, where terminal escape sequences and mouse tracking modes must be faithfully relayed between local and remote systems, this kind of conflict becomes more pronounced and harder to diagnose, since local terminal emulators, SSH clients, and remote shell environments all have to agree on how mouse events are being interpreted.
The practical workaround—simply deleting the `tui: fullscreen` line from settings.json and restarting Claude Code—restores default terminal behavior, suggesting that whatever alternate rendering mode fullscreen TUI enables is not just cosmetic but actively changes how input events are captured and processed by the CLI application. This is a useful data point for both end users and Anthropic's Claude Code development team: it isolates the root cause to a single, well-defined configuration flag rather than a broader SSH or terminal compatibility problem, which should make the underlying bug easier to reproduce and fix upstream.
More broadly, this incident reflects a recurring challenge in the current wave of AI-powered developer tools, many of which are increasingly delivered as terminal-native applications (Claude Code, GitHub Copilot CLI, and similar agentic coding assistants) rather than traditional GUI-based IDEs. As these tools compete to offer richer in-terminal experiences—complete with visual panels, syntax highlighting, and interactive UI elements—they inherit the long-standing complexity of terminal emulation, especially across remote and SSH-based workflows that remain a staple of professional software development. Bugs like this underscore the importance of rigorous cross-environment testing (local vs. remote, different OS/terminal emulator combinations) for CLI-based AI agents, and they illustrate how community-driven troubleshooting on platforms like Reddit and GitHub continues to play an outsized role in surfacing and diagnosing these edge cases faster than internal QA processes alone.
Read original article →