ModelsLab Text2Image API専用MCPサーバー - Flux、Stable Diffusion等の高品質画像生成
This server has 3 tools with inconsistent quality. The main tool (modelslab_text2img) has a well-defined schema with extensive parameters, but the other two are minimal. Naming follows verb patterns adequately. Descriptions vary in quality, modelslab_text2img is detailed (Japanese), but hello and test_connection are sparse. Critical gaps: no output schemas documented, no error recovery guidance, parameter descriptions lack format constraints, and no tool annotations (readOnlyHint/destructiveHint) despite different risk levels.
シンプルなHello Worldレスポンス(テンプレート動作確認用)
ModelsLab APIを使用してテキストプロンプトから画像を生成し、オプションでファイルに保存します
ModelsLab APIへの接続テスト
hello and test_connection tools have extremely sparse descriptions (under 20 characters effectively).
No output schemas documented for any tool. The rubric requires 'Document the output schema. LLMs need to know what fields to expect so they can plan downstream tool calls.' The code uses formatResponse() in ToolBase but never declares what fields clients should expect from hello, test_connection, or modelslab_text2img responses.
Missing tool annotations (readOnlyHint, destructiveHint, idempotentHint). The server marks tools with Risk: READ_ONLY and Risk: WRITE, but these are not reflected in the MCP tool schema. Per current spec (2026-07-28), tools should include tool_annotations via readOnlyHint (for safe tools) or destructiveHint (for writes). modelslab_text2img performs WRITE (image generation + file I/O) but carries no schema annotation.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 43 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 30 | - | v1 |
Parameter descriptions lack actionable constraints. E.g., 'num_inference_steps' has min/max in schema (1 - 50) but description doesn't state range; 'guidance_scale' description mentions 'ガイダンススケール(プロンプトへの従順度)' (guidance scale with adherence) but never specifies 1.0 - 20.0 range or what values achieve good results.
test_connection has an empty input schema ({}) with no parameters, but no description explains what it tests, what prerequisites are needed, or what success/failure looks like. The description '接続テスト' (connection test) doesn't tell an LLM when to call it or how to interpret the result.
Error handling in CallToolRequestSchema catches generic errors (network, validation, timeout) but does not provide recovery guidance. A raw error code or stack trace gives the agent nothing to act on.' Current errors like 'Network connection failed' or 'Invalid arguments format' are minimal; no suggestion for retry, fallback, or next step.
modelslab_text2img accepts 'outputPath' parameter for file I/O but no error handling documents what happens if the directory doesn't exist, write fails, or disk is full. ModelsLabImageHandler.downloadAndSaveImage() handles mkdir recursively, but agent is not told success criteria or failure modes.