Loading ToolBench
MCP server for Akashic game development with tools for documentation search, game project creation, asset extraction, and web scraping
akashic-mcp presents a mixed-quality toolkit with 9 tools spanning API documentation search, project creation, web scraping, and asset extraction. Naming is generally strong (verb-first: search_, get_, create_, list_, extract_, fetch_, take_, generate_). Descriptions are present for all tools and mostly adequate (ranging 50-200 chars, meeting baseline). However, there are significant gaps in schema completeness, parameter descriptions, and output documentation. Parameter descriptions exist but are sometimes generic or lack constraint details. Output schemas are not documented in the source, no indication of what fields are returned, making it difficult for LLMs to plan downstream calls. Web-scraping tools have good security validation (IP blocklisting, credential stripping) but lack error guidance. The toolkit is functional but would benefit from explicit output schema documentation and more detailed parameter constraints.
Create a new Akashic game project with optional template selection and renderer configuration.
Create a distributable ZIP package from an Akashic game project directory.
Extract images and audio assets from a public webpage by URL.
Fetch and convert HTML from a public webpage to markdown format.
Generate a bitmap font (BMP font) from a TrueType or OpenType font file.
Get a summary index of preloaded Akashic API documentation for quick reference.
List all available Akashic project templates.
Output schemas not documented for any tool. LLMs cannot infer what fields to expect in responses, limiting their ability to plan multi-step chains and extract relevant data. For example, search_akashic_docs likely returns an array of results, but the structure (id, title, content, url, relevance_score?) is nowhere specified.
get_api_summary_index and list_templates lack descriptions explaining what they return or when to use them. 'Get a summary index' is vague, does it return names only, or full API signatures? Does it support pagination?
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 52 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 14 | - | v1 |
Search and fetch detailed information from preloaded Akashic docs/repo references.
Take a screenshot of a public webpage and return it as PNG or JPEG.
No input validation errors documented. Tools like fetch_webpage_html and take_page_screenshot perform URL validation in extended-tools.mjs (blocking private IPs, checking protocols), but error responses are not documented in the tool schema. LLMs cannot know whether a failed call is retryable or user-fixable without seeing error responses.
Destructive tools (create_akashic_project, create_akashic_zip_package, generate_bitmap_font) do not declare idempotent or destructive hints. LLMs should know which operations modify state and whether retries are safe. No mention of dry-run or confirmation patterns for create_* tools.
extract_web_assets accepts maxResults up to 100 and assetTypes as a free-form array of enums, but there's no guidance on what happens when the limit is hit. Does it return a 'next_cursor' for pagination? How should the LLM request more results?
Parameter descriptions for several tools are minimal. E.g., search_akashic_docs's 'query' parameter description is just 'Search keyword (e.g. 'click event', 'g.Sprite', 'audio').', examples are embedded in the description rather than constraints being formalized. Per the rubric, examples should be replaced with enum/pattern constraints.