Simple, token-efficient MCP server for Codex and Claude Code image generation with Nano Banana 2 and Nano Banana Pro
Three tools with complete schemas and descriptions. generate_image has comprehensive parameter documentation (14 params, all typed and described). Tool names follow verb_noun pattern. Descriptions are adequate (100-200 chars). Output schemas are explicitly defined with Zod. However, error handling lacks recovery guidance, and some parameter descriptions could be more prescriptive about constraints and dependencies.
Generate or edit an image and save it to IMAGE_OUTPUT_DIR. This tool uses your prompt as-is and does not rewrite or optimize it.
Return the effective non-secret server configuration, including output directory and default quality.
Return a compact readiness report for auth, output-directory access, and effective runtime configuration.
generate_image quality parameter description lacks guidance on model selection. 'fast uses Flash, balanced adds higher thinking, quality uses Nano Banana Pro' is vague about tradeoffs (speed vs cost vs quality). LLMs cannot infer when to choose each.
No error handling guidance in tool descriptions. If Gemini API fails, rate-limits, or rejects a prompt, the LLM has no recovery path. Missing pattern: recovery-guide.
inputImagePath and inputImagePaths are mutually exclusive but descriptions do not state this. LLMs may pass both, causing ambiguous behavior.
image_get_config and image_healthcheck descriptions do not explain when to call each. Both are inspection tools, the distinction is unclear to LLMs.
Inferred effective spec: 2025-06-18+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 68 | 2025-06-18+ | v2 |
No pagination or result limits documented. If generate_image returns large binary data or metadata, no guidance on chunking or streaming.