Loading ToolBench

Reading the current corpus…

nano-banana-mcp MCP server · D · ToolBench

nano-banana-mcp

Free MCP server for AI image generation - no API key, no billing. Uses Nano Banana via Google Flow with your own Google account, at exact pixel sizes, with reference images. $0 per image.

2026-07-28+STDIO5 tools

58/100

Definition Quality73

Protocol Readiness50

Supportability38

Spec posture

Inferred effective spec 2026-07-28+·current baseline 2026-07-28

Static source inference · medium confidence

Current-spec patterns detected

Summary

Server has 5 tools with complete schemas and descriptions. Naming follows verb_noun convention (flow_status, generate_image, download_image, upscale_image, list_library). Descriptions are present and substantive (100-200 chars typical). However, several parameters lack descriptions or have minimal guidance. Output schemas are not documented. Error handling exists but lacks recovery guidance. No tool annotations (readOnlyHint/destructiveHint) despite clear risk profiles.

Tools (5)

download_imagewritesource verified73/100

Download a previously generated image by media ID, with optional resizing, cropping, and upscaling.

flow_statusread onlysource verified77/100

Get the current status of the Flow connection, including browser connectivity, session state, project ID, credit balance, and output directory configuration.

generate_imagewritesource verified75/100

Generate images using Google Flow with support for custom sizes, aspect ratios, reference images, upscaling, and output formatting.

list_libraryread onlysource verified67/100

List all images in the current Flow project library, including uploaded and generated images with their metadata.

upscale_imagewritesource verified73/100

Upscale a previously generated image to a higher resolution (2K or 4K) with optional resizing and cropping.

Issues (5)

HIGHtool ↗

Output schemas not documented. LLMs cannot infer what fields generate_image, download_image, upscale_image, and list_library return. This forces agents to guess at response structure and breaks downstream tool chaining.

generate_imagedownload_imageupscale_imagelist_library

HIGHtool-annotations ↗

Missing tool annotations. generate_image, download_image, and upscale_image are destructive (WRITE risk) but lack destructiveHint annotation. flow_status and list_library are read-only but lack readOnlyHint. Agents cannot distinguish safe from risky operations.

flow_statusgenerate_imagedownload_imageupscale_imagelist_library

MEDIUMtool-description ↗

Parameter 'background' in download_image and upscale_image lacks description. Unclear when/why to use it or what format is expected (hex color, CSS name, RGB tuple?).

download_imageupscale_image

Recommendations

Add tool annotations: server.tool(..., { readOnlyHint: true }) for flow_status and list_library; { destructiveHint: true } for generate_image, download_image, upscale_image.
Document output schemas for all tools. E.g., generate_image returns { images: [{mediaId, signedUrl, width, height}], quotedCost: number }.
Add descriptions to 'background' parameter: 'CSS color name or hex code (e.g., white, #ffffff) for contain fit mode background.'
Enhance error messages with recovery hints. E.g., fail() should return 'Invalid size format. Expected WIDTHxHEIGHT (e.g., 1200x630). Got: ' + args.size.
Document list_library output: 'Returns array of library items with fields: id, name, type, uploadedAt, dimensions. Limit: 100 items per call.'

Score history

Overall score trend

First recorded score · v2 rubric

58/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-23	D	58	2026-07-28+	v2