Voice MCP Server has three tools with explicitly defined schemas and descriptions. All tools have clear action verbs and reasonable parameter descriptions. However, there are notable gaps: (1) Output schemas are not documented, only input schemas are visible. (2) Tool descriptions lack detail about when to use each tool vs. alternatives and do not explicitly state which operations modify state. (3) Some parameter descriptions are thin and lack constraints or format guidance. (4) Error handling is minimal, no recovery guidance or error classification visible. (5) The 'listen' and 'action' tools are READ_ONLY but lack explicit idempotent/readonly hints in the schema. The server demonstrates competent basic schema definition but falls short of production-grade clarity and error handling expected at 70+.

• Document the output schema for each tool. For 'speak', specify: { ok: boolean, message: string, messages?: string[] }. For 'listen', specify: { messages: string[] }. For 'action', specify the return type (e.g., { ok: boolean }).
• Replace examples in the 'action' tool description with constraint text: 'Concrete, specific action being performed, typically 5 - 100 words (e.g., 'Reading useWebSocket.ts' or 'Running build command'). Use simple present tense and avoid explanations, just state what the tool is doing now.'
• Clarify the relationship between 'speak(listen=true)' and the standalone 'listen' tool. Suggest: 'Use speak(listen=true) to ask a question and immediately wait for the user's response in one call. Use listen() only when you need to wait for input asynchronously after other work.'
• Add error recovery guidance to tool descriptions. For 'speak': 'If the client is not connected, you cannot stream audio. Inform the user that voice interaction requires an active connection.' For 'listen': 'A timeout (no response within the specified time) returns an empty messages array. The user may not be actively responding; consider rephrasing your request or moving forward with a default action.'
• Specify constraints for all parameters. Examples: 'text: 1 - 2000 characters (ElevenLabs TTS limit); timeout_ms: 1 - 300000 (1ms to 5 minutes); voiceId: ElevenLabs voice ID string (defaults to 'rachel' if omitted); modelId: one of 'eleven_flash_v2_5' or 'eleven_multilingual_v2' (defaults to 'eleven_flash_v2_5' if omitted).'