Loading ToolBench
An MCP server that exposes tmux operations across a process boundary using LibTmux, providing tools for inspecting, managing, executing commands, and tearing down tmux sessions, windows, and panes.
This server has 45 tools with highly inconsistent quality. Only 3 tools (capture_since, search_panes, create_session, show_option, paste_text) have visible descriptions and input schemas in the source. The remaining 40 tools are defined in ToolOperation.swift but their schemas and descriptions are not visible in the provided code, they appear to be inferred from file references only. Per the rubric, tools without visible definitions are capped at 50. Most parameter descriptions are missing entirely. Error handling is minimal. The server does expose tool annotations (readOnlyHint/destructiveHint), which is a positive signal for current spec alignment, but definition quality is severely hampered by incomplete schema visibility and missing descriptions across the majority of tools.
Capture pane output since a previous cursor position, with optional wait for new output
Create a new tmux session with optional window configuration
40 of 45 tools lack visible input schemas and descriptions in provided source code. Only 5 tools (capture_since, search_panes, create_session, show_option, paste_text) have documented schemas and descriptions.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 33 | 2025-06-18+ | v2 |
Paste text into a pane using tmux paste buffer mechanism
Search pane contents for a pattern across multiple panes with work budgets and progress reporting
Show tmux options for a given scope and optional target
No error handling guidance visible. Tools lack recovery hints (e.g., 'If pane not found, try list_panes first'). Error responses do not categorize as retryable, user-fixable, or fatal.
Destructive tools (kill_pane, kill_session, kill_window, clear_pane_scrollback) lack confirmation or dry-run support. No pattern to prevent accidental data loss.
Parameter descriptions missing for 40 tools. LLMs cannot infer what 'target', 'scope', 'pattern', or 'regex' mean without explicit descriptions. Baseline: 100% of A+ tools have param descriptions.
Output schemas not documented. LLMs cannot plan downstream tool calls or extract required fields (e.g., pane_id from list_panes) without knowing response structure.