Loading ToolBench
MCP server for image generation and editing via Google Gemini API (Nano Banana 2). Provides tools for text-to-image generation, image editing, technical diagram creation, and icon generation.
Nano Banana MCP server demonstrates solid definition quality with well-structured tool schemas, comprehensive parameter documentation, and clear descriptions. All four tools (nanobanana_generate, nanobanana_edit, nanobanana_icon, nanobanana_diagram) have properly typed input schemas with enum constraints and parameter descriptions. However, there are gaps in output schema documentation and error handling guidance that prevent a higher score. The tool names are action-oriented (generate, edit, icon, diagram) and follow verb_noun conventions. Descriptions average 250-400 characters, exceeding the production baseline of 194 chars, with detailed capability documentation. Parameters consistently include type definitions, enums for constrained inputs, and descriptions explaining purpose and format. No security concerns detected, API keys use environment variable injection, not parameters. The main limitation is lack of documented output schemas and error recovery guidance.
Generate professional technical diagrams from text descriptions. Creates publication-ready diagrams with specialized prompt engineering for each type: - flowchart: Process flows with decisions, actions, and connections - architecture: System/software architecture with components and data flow - database: ER diagrams with tables, fields, and relationships - sequence: Interaction diagrams between actors/systems over time - mindmap: Hierarchical concept maps with branching ideas Available styles: professional, minimal, colorful, blueprint Complexity levels: simple (3-6 elements), standard (6-12), detailed (12+) Example usage: - prompt: "user authentication flow with OAuth2", type: "flowchart" - prompt: "microservices e-commerce platform", type: "architecture", complexity: "detailed" - prompt: "social media database schema", type: "database", style: "blueprint"
Edit an existing image using natural language instructions. Provide the path to a source image and describe the changes you want. The edited image is saved as a new PNG file. Capabilities: - Add, remove, or modify elements - Change backgrounds, colors, or lighting - Apply style transfers - Restore or enhance old photos - Remove unwanted elements Example usage: - image_path: "photo.jpg", prompt: "Add sunglasses to the person" - image_path: "landscape.png", prompt: "Change to a sunset sky with warm colors" - image_path: "old_photo.jpg", prompt: "Remove scratches, enhance clarity, restore colors"
Generate an image from a text prompt using Nano Banana (Gemini image models). Supports advanced options: aspect ratio, resolution, style, and model selection. The generated image is saved as PNG to the specified output directory. Available aspect ratios: 1:1, 1:4, 1:8, 2:3, 3:2, 3:4, 4:1, 4:3, 4:5, 5:4, 8:1, 9:16, 16:9, 21:9 Available resolutions: 512px, 1K, 2K, 4K Available models: gemini-3.1-flash-image-preview (Nano Banana 2, default), gemini-3-pro-image-preview (Nano Banana Pro) Example prompts: - "A watercolor painting of a fox in a snowy forest" - "A professional product photo of a coffee cup on a marble table" - "An isometric 3D cartoon scene of Tokyo with cherry blossoms"
Output schemas are not documented. Tools describe what they generate (PNG files to disk) but do not specify the structure of the return object. LLMs cannot plan downstream operations without knowing what fields are returned.
Error handling lacks recovery guidance. The code may throw errors (e.g., 'Source image not found', missing API key, invalid paths) but descriptions do not tell the LLM what to do next or how to retry. Errors should be actionable.
nanobanana_edit requires an image_path parameter but does not document accepted formats (JPEG, PNG, WebP, etc.) or maximum file size. LLMs may pass unsupported formats, causing failures without guidance on what went wrong.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | C | 69 | 2026-07-28+ | v2 |
| 2026-03-09 | D | 58 | - | v1 |
Generate app icons, favicons, and UI elements with specialized prompt engineering. Creates icons with proper formatting for each type: - app-icon: iOS/Android app icons with clean backgrounds - favicon: Ultra-simple designs optimized for 16-32px - ui-element: Interface icons for modern applications Each icon is generated at multiple sizes (default: 64, 128, 256px). Specialized prompts ensure consistent, professional results. Available styles: minimal, flat, 3d, outline, gradient, glassmorphism Example usage: - prompt: "coffee cup logo", type: "app-icon", sizes: [128, 256, 512] - prompt: "lightning bolt", type: "favicon", sizes: [16, 32, 64] - prompt: "settings gear", type: "ui-element", style: "outline"
Parameter constraints in descriptions use informal language ('e.g. watercolor, oil-painting, pixel-art') rather than enums. The 'style' parameter in nanobanana_generate accepts free-form strings instead of constrained options, inviting hallucinated values.
output_dir and filename parameters default to './' and auto-generated values respectively, but there is no documented output structure. Users cannot reliably locate generated files or understand the naming convention without examining source code.