MCP server: Claude-rendered images/video + BYOK photoreal generation
Server has 5 tools with complete input schemas and descriptions. Naming follows verb_noun convention (render_*, generate_*, list_*). Descriptions are adequate (50-150 chars) but lack actionable context for LLM selection. Parameters are typed with enums for constrained inputs (format, output_format, aspect_ratio). Critical gap: no output schema documentation, LLMs cannot predict what fields to extract from responses. Error handling is minimal; no recovery guidance. Security is strong (API keys injected via env, not params). Composition is clean (single responsibility per tool). Baseline: 549 production tools average 194 chars descriptions; these average ~90 chars, below median.
Generate photorealistic images using configured AI provider (OpenAI, Google Gemini, or Fal)
Generate videos using configured AI provider (OpenAI Sora, Google Veo, or Fal Kling)
List available media generation providers and their capabilities
Render SVG or HTML to PNG/JPEG/WebP image
Render animated video from JavaScript draw() function
No output schema documentation. LLMs cannot predict response structure (fields, types, pagination). Responses must document what fields are returned so agents can chain tools and extract data reliably.
Descriptions are terse (50-90 chars, below 194-char baseline). Missing WHEN to use each tool and dependencies. E.g., generate_image lacks guidance on when to use vs render_image, or what 'configured AI provider' means to an LLM unfamiliar with the server.
No error handling guidance. Tools return isError=true but no recovery hints. E.g., if generate_image fails due to invalid prompt or API quota, LLM has no actionable next step. Should include 'Try list_providers() to verify API keys are configured' or similar.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 57 | 2026-07-28+ | v2 |
Parameter descriptions lack format/constraint details. E.g., 'prompt' in generate_image has no guidance on length, style, or forbidden content. 'duration_seconds' in generate_video lacks min/max bounds. LLMs will guess and pass invalid values.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). render_image and render_video write files but lack destructiveHint. list_providers is read-only but lacks readOnlyHint. Annotations help agents reason about side effects and retry safety.