Smart JSON-to-TOON gateway with pragmatic auto-gating for LLM prompts.
Three tools with complete input schemas and reasonable descriptions. All tools have verb-noun naming (convert_*, analyze_*) and clear parameter types. However, descriptions are generic and lack LLM-optimized guidance on WHEN to use each tool vs. alternatives. Output schemas are documented as dicts but lack field-level descriptions. Error handling returns error strings but provides no recovery guidance. Parameter descriptions are minimal (e.g., 'JSON payload to convert' lacks format/size constraints). No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite all being read-only operations.
Analyze a JSON payload and report whether it is a TOON candidate. Does not invoke the TOON CLI — safe to call without Node.js installed.
Convert a JSON payload to TOON when structure and savings meet policy. Returns the final payload text (TOON or original JSON) and a report explaining the conversion decision.
Convert structured text (csv, yaml, xml, jsonl) to TOON with policy gating. Supported formats: csv, yaml, xml, jsonl. For JSON input use convert_json. Binary formats (parquet, excel, etc.) require the CLI.
Output schemas undocumented. Tools return dict[str, Any] with fields like 'payload', 'report', 'error' but no schema definition visible to LLMs. LLMs cannot plan downstream operations or extract structured data reliably.
Parameter descriptions lack actionable constraints. 'JSON payload to convert' does not specify max size, nesting depth limits, or format requirements. 'Minimum savings ratio threshold' lacks range (0.0 - 1.0?) or units. LLMs cannot validate inputs before calling.
Tool descriptions lack WHEN-to-use guidance. No explanation of when to call convert_json vs. convert_text, or when analyze_json is preferable to convert_json. LLMs must guess based on parameter names alone.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 60 | <=2025-11-25 | v2 |
Error handling provides no recovery guidance. Errors return bare strings like 'Invalid JSON: ...' or 'Unsupported format'. No suggestion of next steps (e.g., 'Try analyze_json first to validate structure').
No tool annotations despite all tools being read-only. FastMCP supports readOnlyHint but none are declared. This prevents clients from optimizing caching or safety policies.