Textile • Modular AI Subsystem & Desktop Intelligence Engine. A Python-based MCP server providing desktop automation, virtual input, and system interaction capabilities through a layered yarn/strand architecture.
Textile exposes three desktop automation tools with basic descriptions but significant quality gaps. All three tools (type_text, press_key, paste_text) have descriptions present (10-100 chars range), but parameter documentation is sparse. Input schemas are visible in the code with type hints (string, boolean, integer), but lack formal JSON Schema constraints (enums, min/max, patterns). No output schemas are documented. Error handling returns plain strings rather than structured guidance. The tools perform write operations (WRITE risk) but lack confirmation patterns or dry-run support. Tool names follow verb_noun convention (positive), but descriptions are terse and lack WHEN/WHY context. No per-parameter descriptions visible in the schema definitions, only in docstrings.
Paste text or clipboard content into the active focused window.
Press a specific keyboard key or hotkey combination.
Type text, commands, or passwords directly into active focused window or terminal.
No JSON Schema constraints (enums, min/max, patterns) on parameters. delay_ms accepts 1-200 but no bounds declared; key accepts arbitrary strings with no enum of valid keys despite KEY_MAP existing in code.
No output schemas documented. Tools return plain strings ('Successfully typed text into active window.') with no structured response format. LLMs cannot extract success/failure state or downstream references.
Destructive operations (type_text, paste_text, press_key) lack confirmation or dry-run support. No confirmation_request pattern. Agents cannot preview actions before executing irreversible desktop changes.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 49 | <=2025-11-25 | v2 |
Parameter descriptions missing from schema definitions. Only docstrings document params; the @strand decorator and schema lack per-parameter descriptions. LLMs see only type names (text, key, modifiers) without context.
Error responses are unstructured strings with no recovery guidance. 'Error: Virtual keyboard daemon (ydotool) could not connect.' gives no actionable next step. No error classification (retryable vs fatal).