Translate text and documents, extract text from images, transcribe audio, synthesize speech, query glossaries, and list supported languages via LLM-powered translation.
Server provides 7 tools with complete descriptions and mostly well-structured schemas. Tool names follow verb_noun convention (translate_text, extract_image_text, etc.). All tools have descriptions (avg ~150 chars, within 10-1024 baseline). Input schemas are present with type declarations and parameter descriptions. However, output schemas are not documented, LLMs cannot predict return structure for chaining. Error handling is minimal; no recovery guidance or actionable error messages visible. Parameters lack enums where appropriate (e.g., content_type accepts free-form strings instead of constrained values). No tool annotations (readOnlyHint/destructiveHint) despite clear risk levels. translate_document and synthesize_speech are WRITE operations but lack confirmation/dry-run patterns.
Extract text from an image using OCR or the configured LLM vision model. Tries the LLM vision provider first (Gemini / custom endpoint). Falls back to OCR when LLM isn't configured OR when LLM returns empty/whitespace text. LLM errors (auth, quota, network) propagate as-is rather than silently falling back — otherwise misconfiguration would be invisible to the caller.
List all supported languages for translation.
Query a glossary for term translations or definitions.
Synthesize speech from text using the configured text-to-speech engine.
Transcribe audio from a file or microphone input using the configured speech-to-text engine.
Translate a document file (PDF, Word, Excel, PowerPoint, etc.) into the target language.
Output schemas not documented. LLMs cannot predict return structure (fields, types, pagination) for downstream tool chaining or data extraction. Breaks pattern:tool-chain and pattern:response-shaper.
content_type parameter in translate_text accepts free-form strings instead of enum constraint. Description lists examples ('plain_text', 'html', etc.) but LLMs may hallucinate invalid values. Should be enum: [plain_text, html, subtitle, markdown, xml, rtf, json, localization].
WRITE operations (translate_document, synthesize_speech) lack confirmation/dry-run pattern. Agents may accidentally overwrite files or generate unwanted audio without user approval. No idempotency guarantees documented.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 58 | 2026-07-28+ | v2 |
Translate a list of text strings into the target language.
Tool annotations missing. FastMCP supports readOnlyHint and destructiveHint but none are declared. Clients cannot distinguish safe (list_languages, query_glossary) from risky (translate_document) operations.
Error handling not visible in source. No recovery guidance (e.g., 'If model not found, call list_models()'), no error categorization (retryable vs user-fixable), no actionable error messages. Breaks pattern:recovery-guide.