Server has 5 tools with complete schemas and descriptions. Naming is clear and verb-based (transcribe_*, list_*, check_*, speak_*). Descriptions are detailed (100-300 chars) and explain WHAT, WHEN, and HOW. Parameters have types and descriptions. However, output schemas are not formally documented in the code, only inferred from docstrings. Error handling returns structured dicts but lacks recovery guidance (e.g., 'try X next'). The speak_text tool is WRITE-risk but lacks confirmation/dry-run pattern. No tool annotations (readOnlyHint/destructiveHint) visible in FastMCP registration.

• Add formal JSON Schema output definitions to each tool. Document the structure of returned dicts (e.g., {text: string, language: string, segments: [{start: float, end: float, text: string}], success: bool, error: string}).
• Enhance error messages with recovery hints. E.g., 'File not found: /path/to/file. Ensure the path is absolute and the file exists.' or 'No bot token provided. Set TELEGRAM_BOT_TOKEN env var or pass bot_token parameter.'
• Add tool annotations to FastMCP decorators: @mcp.tool(annotations={'readOnlyHint': True}) for transcribe_* and list_* tools; @mcp.tool(annotations={'destructiveHint': True}) for speak_text.
• For speak_text, implement a dry-run mode (e.g., dry_run=True parameter) that returns the TTS backend and estimated cost without writing the file.
• Clarify discovery tool usage in descriptions. E.g., 'Call check_backends() first to verify Whisper and TTS backends are available before transcribing.'

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry