Loading ToolBench
Professional RAG-MCP: Advanced Danbooru Prompt Generation System with BGE-M3 Triple Vector Search for semantic tag matching, artist search, and intelligent prompt optimization
Server has 9 tools with READ_ONLY risk classification. Tool names follow verb_noun pattern (search_tags, analyze_prompts, create_prompt_from_scene, optimize_prompt, search_artist, get_performance_metrics, detect_query_intent, apply_tag_aliases, detect_nsfw_level). Descriptions are present and moderately detailed (50-150 chars typical). However, critical gaps exist: input schemas are not visible in the provided source code, only parameter names and types are listed in the summary, not actual JSON Schema definitions. Output schemas are completely undocumented. Error handling is absent from all tool definitions. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite READ_ONLY classification. Parameter descriptions exist but lack constraints (enums, ranges, patterns). The code snippet shows imports and configuration but does not show actual tool registration with @mcp.tool() decorators or explicit schema definitions. This forces a conservative score: visible descriptions prevent a failing grade, but missing schemas and error guidance are critical gaps.
Analyze and validate multiple prompts for quality, completeness, and Danbooru tag compliance
Apply Danbooru tag alias mappings to normalize and standardize tag names, fixing common misspellings and variations
Generate optimized Danbooru prompts from natural language scene descriptions with automatic tag decomposition and quality enhancement
Detect NSFW content level in text based on indicator keywords and return risk assessment
Intelligently detect the intent of a search query (artist, nsfw, character, copyright, appearance, pose, expression, or general)
Retrieve server performance metrics including query statistics, cache hit rates, response times, and success rates
Intelligently optimize raw prompts by adding quality tags, fixing tag aliases, and enhancing semantic meaning based on detected intent
Input schemas not visible in source code. Parameter types and descriptions are listed in summary but actual JSON Schema definitions with type, properties, required arrays are not shown in danbooru_prompt_server_v2_minimal.py. Cannot verify schema compliance.
Output schemas completely undocumented. No tool returns a documented response structure. LLMs cannot plan downstream calls or extract fields reliably.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 42 | <=2025-11-25 | v2 |
Search for artists and their signature styles, techniques, and characteristic elements in Danbooru database
Search for Danbooru tags using BGE-M3 triple vector search (dense, sparse, and ColBERT vectors) with intelligent query enhancement and NSFW detection
No error handling guidance. Tools lack recovery instructions, error categorization (retryable vs fatal), or actionable error messages. Agents cannot self-correct on failures.
No tool annotations despite READ_ONLY classification. Tools should declare readOnlyHint=true to signal safety to agents. Missing destructiveHint and idempotentHint annotations.
Parameter constraints missing. 'limit' parameters lack min/max bounds (spec says default 20, max 200 but no JSON Schema minValue/maxValue). 'intent' and 'nsfw_level' should be enums, not free-form strings.