store: Decision: mcp-tui-driver chosen over mcpterm for TUI testing

graph/decisions/decision-mcp-tui-driver-chosen-over-mcpterm-for-tui-testing-230a1d.md

@@ -0,0 +1,43 @@
+---
+id: 230a1d01-d47b-46ee-91f1-7388e3255b19
+type: decision
+title: "Decision: mcp-tui-driver chosen over mcpterm for TUI testing"
+tags: [mcp, tui-testing, architecture-decision, claude-code, mcpterm, claude-home]
+importance: 0.7
+confidence: 0.8
+created: "2026-03-02T00:38:13.416653+00:00"
+updated: "2026-03-02T00:38:13.416653+00:00"
+---
+
+# Architecture Decision: TUI Testing MCP Server Selection
+
+## Context
+
+Needed an MCP server to enable Claude Code to automate and test TUI (terminal UI) applications.
+
+## Options Evaluated
+
+### mcpterm (Go)
+- Only 2 tools
+- Proof-of-concept quality
+- Single session only
+- Minimal feature set
+
+### mcp-tui-driver (Rust)
+- 23 tools
+- Playwright-like design philosophy
+- Multi-session support
+- PNG screenshots for visual verification
+- Accessibility snapshots with element references (structured, not just raw text)
+- Mouse support (click, double-click, right-click, coordinate-based)
+- Proper wait primitives (`tui_wait_for_text`, `tui_wait_for_idle`)
+- Asciicast recording
+- JS scripting via `tui_run_code`
+
+## Decision
+
+Chose **mcp-tui-driver**. The accessibility snapshot approach (element references rather than raw terminal text) and proper wait primitives make systematic test development practical. mcpterm is too minimal for anything beyond simple PoC work.
+
+## Source
+
+https://github.com/michaellee8/mcp-tui-driver