Loading ToolBench
The anti-AI-slop standard as an MCP server: your AI agent reads the standard, checks text against a rule engine, and gets a rewrite prompt.
Four tools with clear, domain-specific descriptions (194 - 280 chars, within baseline 34 - 392). All tools are read-only and properly scoped. Schemas present for 3/4 tools with Zod validation. However, antislop_overview has an empty input schema ({}), and parameter descriptions lack depth on constraints/formats. No error handling guidance, no output schema documentation, and no tool annotations (readOnlyHint). Naming is verb-forward (antislop_*) but generic; descriptions are good but could better explain when/why to call each tool vs. alternatives.
Check a piece of text against the anti-AI-slop rule engine. Returns the mechanical tells found (symbols, banned vocabulary, contrast and binary framing, false agency, sycophancy, Indonesian calques and marketing cliches), an editorial style score (0 to 100, higher is cleaner), and a ready-to-paste prompt to rewrite the text to natural. The score is an editorial style signal, not an authorship verdict.
List the engine's rules in one category, by category id (for example symbol, vocabulary, semantic, false-agency, sycophancy, calque, marketing). Each rule shows its severity, what it flags, and the fix.
List the anti-AI-slop pattern categories, each with a one-line note and how many rules the engine checks in it.
Return the anti-AI-slop standard spine: what slop is, the remediation discipline (detect then rewrite to natural), the pattern categories (symbols, vocabulary, contrast and binary framing, false agency, cadence, sycophancy, the 2026 model-tell panel, Indonesian-specific tells), and the detection-claims discipline. Read this first before writing or editing text that must not read as AI-generated.
Two tools (antislop_overview, antislop_list_categories) have empty input schemas ({}). These tools should document their output structure (e.g., 'Returns text content with categories and rule counts').
No output schema documentation. Tools return text content but LLMs cannot infer structure. antislop_check returns findings array, score, and rewrite prompt, document this so agents can parse and chain results.
No error handling guidance. If antislop_explain receives an invalid category, it returns a plain text error. Should guide LLM: 'Invalid category. Call antislop_list_categories to see valid options.'
Parameter descriptions lack constraint details. 'lang' enum is documented but no note on default behavior ('both' if omitted). 'severityMin' lacks guidance on what 'low', 'med', 'high' mean in context.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 54 | 2026-07-28+ | v2 |
No tool annotations. All four tools are read-only (no side effects), but readOnlyHint is not set. This prevents agents from reasoning about safety and retry logic.