Two tools with complete input schemas and descriptions, but significant gaps in output documentation and error guidance. Tool names follow verb_noun pattern (query_ollama, query_ollama_with_files). Descriptions are adequate (100-150 chars) but lack WHEN/WHY context. Parameters are typed and described, but output schemas are not documented, LLMs cannot plan downstream calls or extract structured data. Error handling returns strings rather than structured errors with recovery guidance. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite both tools being read-only. The second tool relies on deprecated Roots feature for file access, which is a significant architectural issue.

• Document output schemas for both tools. query_ollama should return {response: string, model: string, tokens_used?: number}. query_ollama_with_files should return {results: [{file: string, response: string, status: 'success'|'error', error?: string}], total_files: number, processed: number}.
• Add readOnlyHint annotation to both tools since they do not modify state. This signals to clients that results can be cached and calls are safe to retry.
• Enhance error messages with recovery guidance. Replace 'Error: Model not found' with 'Model "llama3" not found. Available models: mistral, neural-chat. Call query_ollama with one of these model names.'
• Expand tool descriptions to include prerequisites and use cases. E.g., 'Query a local Ollama model for inference. Requires Ollama server running on localhost:11434. Use this for testing prompts against small models before deploying to production LLMs.'
• Migrate away from Roots-based file access. Accept file_path as a direct parameter with validation, or document how to configure roots via server initialization rather than relying on deprecated RootsState.

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

• Relies on Roots (deprecated) - pass files/dirs via tool parameters, resource URIs, or configuration