Loading ToolBench
MCP server that searches booru posts and cleans tags into ready-to-use AI art prompts (Danbooru/Gelbooru/e621/AIBooru/Rule34)
Strong tool definitions with comprehensive descriptions and well-structured schemas. All 9 tools have detailed descriptions (100-500+ chars) explaining when to use them and how they differ from alternatives. Input schemas are complete with types, enums, and parameter descriptions. However, output schemas are not explicitly documented in the source, and error handling guidance is minimal. Tool naming is clear and verb-based (search_, build_, get_, autocomplete_). The server demonstrates good composition, tools are single-purpose and chainable. Main gaps: no documented output structure, no error recovery guidance, and some parameter descriptions could be more concise.
Booru tag autocomplete with alias/fuzzy/Chinese-Japanese other_names matching and auto-correction of misspellings. Resolves tag names and provides suggestions.
Clean an already-provided tag set (no network). Takes raw booru tags and returns a cleaned, ready-to-use AI art prompt with category breakdown.
Full character profile in one call (visual trait frequencies + wiki + bidirectional tag implications). Returns comprehensive character information including appearance traits, wiki text, and tag relationships.
Tag implication chain (A -> B auto-adds). Returns forward and backward tag implications showing which tags automatically add other tags.
Textual wiki page for a tag. Returns the Danbooru wiki documentation for a given tag.
Find/verify a tag's canonical name. Resolves aliases and misspellings to the official Danbooru tag name.
Output schemas not documented. Tools return structured data (prompt, categories, raw_tag_count, etc.) but the response structure is not formally specified in tool definitions. LLMs cannot plan downstream operations without knowing what fields to expect.
Error handling lacks recovery guidance. No tool descriptions explain what to do if a search returns no results, a tag is not found, or a provider API fails. Errors should guide the LLM to alternative tools or corrective actions.
Pagination not explicitly documented in tool descriptions. search_prompts and search_posts accept page/limit but descriptions do not state max results per page, total count behavior, or how to iterate. Agents may not know pagination is available.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 65 | 2026-07-28+ | v2 |
Visual trait co-occurrence frequencies for a character tag. Returns a frequency table of appearance traits commonly associated with the character.
Raw post list (no cleaning). Search a booru and return unprocessed post data with all tag categories.
Search an image booru (Danbooru/AIBooru/e621/Gelbooru/Rule34) by TAG and return ready-to-use, cleaned AI-art prompts. Each result has the full prompt string plus its category split (appearance/clothing/pose/scenery/character/quality/other). This is the recommended tool for finding 'ready-to-use AI painting tags'. == WHEN TO USE A DIFFERENT TOOL INSTEAD == - 'What does character X look like?' / 'list costume variants of X' / 'what tags describe X' -> use `danbooru_get_character_profile` (one call: trait frequency table + wiki + bidirectional implications). It is built for *describing a known character*, while `search_prompts` returns *sample images each turned into a prompt* — different goals. - 'Find the canonical tag name for a Chinese/Japanese term or a misspelling' -> use `autocomplete_tags` (alias/fuzzy/other_names aware). == SEARCH SEMANTICS — read this before calling == - This is NOT a keyword-search engine like Google. You query with one or more booru TAGS, and results are images ALL of whose tags are present. - A booru tag is a single token, multi-word tags use UNDERSCORE: 'blue_hair', 'hatsune_miku', 'sitting_on_chair' — never 'blue hair'. - MULTI-TAG SEARCH IS SUPPORTED but each provider caps how many plain tags you may combine per query: danbooru/aibooru/e621 = 2, gelbooru/rule34 = 10 (metatags like order:rank or rating:safe do not count). Exceeding the limit returns an error telling you how to fix it; you can also raise a provider's cap with the BOORU_MAX_TAGS_<PROVIDER> env var (e.g. BOORU_MAX_TAGS_DANBOORU=6 for a Gold account). - STRONGLY PREFER A SINGLE TAG to start. 'hatsune_miku' returns thousands of well-tagged images; 'hatsune_miku blue_hair school_uniform smile' is treated as AND → dramatically fewer results (often zero). The cleaned prompt already contains far more detail than your search tag, so adding search tags is at best redundant and at worst returns nothing. - When you need a pure-quality / generalized set (no specific character or franchise yet), search '1girl' or '1boy' alone, or use 'random' order. - Do not write natural-language queries ('a girl with blue hair sitting in a classroom') — that is not supported. Translate to booru tags first. - Use `autocomplete_tags` if you are not sure how a concept is spelled as a booru tag (e.g. user said '法国女仆' → try 'french', 'maid' and let autocomplete return 'french maid', 'maid_uniform', etc.). - Common useful single-tag starting points: a character tag ('hatsune_miku'), an artist tag, a copyright tag ('blue_archive'), a character + single qualifier ('2girls' for group shots), or 'order:rank' / 'random' for non-themed browsing via `order` param.
Parameter descriptions for background_mode and simple_background_replacement_tags are minimal. The enum values (keep, remove, replace, simple_random, detailed_random) lack explanation of what each mode does. LLMs cannot reason about which mode to select.
Danbooru-specific tools (danbooru_*) lack context on when to use them vs multi-provider tools. Descriptions do not explain that these are Danbooru-only and cannot be used for other boorus, which may confuse agents.