Critical failures across all dimensions. The single tool 'local_llm_agent' has no visible input schema, no parameter descriptions, and a generic 1-line description (28 chars). Source code shows only package.json metadata; the actual tool implementation in 'mcp-servers/local-agent/index.js' is not provided, making schema verification impossible. Tool definition appears inferred rather than explicitly registered. No error handling, no output schema documentation, no security considerations visible.

• Provide the complete index.js implementation file showing explicit tool registration with @modelcontextprotocol/sdk. Include tool name, description (50-200 chars), input schema with typed parameters, and output schema.
• Expand tool description to 50-200 characters explaining: (1) What the tool does (invokes a local LLM via Ollama), (2) When to use it (for local inference without API calls), (3) What it returns (structured response with model output, tokens, latency).
• Define input schema with explicit parameter types and descriptions. Example: { 'prompt': { 'type': 'string', 'description': 'The text prompt to send to the local LLM model' }, 'model': { 'type': 'string', 'description': 'Ollama model name (e.g., llama2, mistral)' }, 'temperature': { 'type': 'number', 'minimum': 0, 'maximum': 2, 'description': 'Sampling temperature (0=deterministic, 2=max randomness)' } }.
• Document output schema with all returned fields. Example: { 'response': string, 'model': string, 'tokens_used': number, 'latency_ms': number, 'stop_reason': string }.
• Add error handling with recovery guidance. Example: If Ollama is unreachable, return 'Ollama service not running on localhost:11434. Start it with: ollama serve' instead of a raw connection error.