mcp 0.5.0 tool output can contain escape sequences that mess with terminal display
Tested MCP 0.5.0 with tool that returns output containing ANSI escape sequences for terminal control. Tool output included `\x1b[2J\x1b[H` (clear screen and move cursor) and it passed straight through to agent's display context, which caused terminal to clear when output was displayed. This means tool output can: 1. Clear user's terminal screen 2. Reposition cursor to overwrite previous output 3. Change terminal colors/formatting in ways that persist 4. Potentially hide malicious output by moving it off-screen SDK should sanitize or escape ANSI control sequences in tool output before passing to display. Tested on SDK 0.5.0, terminal was iTerm2 on macOS but this would affect any ANSI-compatible terminal. Is there any plan to add output sanitization or is expectation that tool implementers manually strip control sequences?
I tested this with ANSI escape sequences in tool output on 0.5.0 - tool returned `\x1b[2J\x1b[H` (screen clear) and it executed in terminal display, completely messed up the agent output. SDK passes tool output straight through with zero sanitization
saving this.... we need to strip ansi escapes from all tool output before it hits the agent or terminal display
iirc this is a terminal emulation issue not sdk issue but yeah stripping ansi escapes before display makes sense
stripping ansi escapes before display makes sense