Loading ToolBench
MCP server that proxies calls to the Google Stitch API for AI-powered UI/UX design generation and management
Stitch MCP has 12 tools with consistent naming (verb_noun pattern) and descriptions present for all tools and parameters. However, schemas lack formal JSON Schema type declarations and enums for constrained inputs. Parameter descriptions are adequate but generic in places. Output schemas are not documented. Error handling is minimal, most tools return raw API responses without recovery guidance. The server acts as a proxy to a remote Stitch API, which limits its ability to provide rich error context and validation.
Apply a design system to screens.
Create a new design system with foundational design tokens.
Create a new Stitch project.
Edit existing screens using a text prompt. Takes a few minutes to complete. Do NOT retry immediately on connection errors — use get_screen after a few minutes.
Generate a new screen from a text prompt. Takes a few minutes to complete. Do NOT retry immediately on connection errors — use get_screen after a few minutes.
Generate design variants of existing screens.
No formal JSON Schema type declarations. Parameters documented in docstrings but not in machine-readable schema format. LLMs cannot validate inputs against declared types/enums before calling.
Enum values documented in descriptions (e.g. 'MOBILE, DESKTOP, TABLET, AGNOSTIC') but not declared as JSON Schema enums. LLMs may hallucinate invalid values like 'LAPTOP' or 'PHONE'.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 52 | 2025-06-18+ | v2 |
Retrieve details for a single Stitch project.
Retrieve details for a specific screen, including download URLs for htmlCode, screenshot, and figmaExport.
List all design systems for a project, or global design systems if no projectId.
List all Stitch projects accessible to the user.
List all screens within a Stitch project.
Update an existing design system.
Output schemas not documented. Tools return raw API responses without specifying field types, structure, or what data is available. Agents cannot plan downstream calls or extract required fields reliably.
Error handling is minimal. call_stitch() returns raw API errors or generic connection messages without recovery guidance. LLMs cannot determine if errors are retryable, user-fixable, or fatal.
Long-running operations (generate_screen_from_text, edit_screens, generate_variants) timeout after 300s with generic 'Request timed out' message. No guidance on polling strategy, expected completion time, or how to check status.