This server presents a single tool 'select_relevant_tools' that is severely underdeveloped. The tool definition is not visible in the provided source code, only a re-export from server.adapters is shown in server/nlp/tool_selector.py. Without access to the actual tool registration (input schema, full description, parameter definitions), the tool cannot be properly evaluated. The description 'NLP tool-selection entrypoint for selecting relevant tools based on user input' is vague (75 chars) and lacks specificity about what 'tool selection' means, when to call it, or what it returns. No input schema is visible, no parameter documentation, no output structure defined. The re-export pattern suggests the implementation is hidden behind adapters, making it impossible to verify schema quality, error handling, or composition patterns. This violates fundamental MCP tool definition practices.

• Expose the full tool definition from server.adapters. Currently only a re-export is visible, provide the actual registration code with input schema, parameter descriptions, and output structure.
• Create a detailed input schema with JSON Schema type definitions. At minimum: parameter name, type (string|array|object), description, and constraints (enum, minLength, pattern, etc.).
• Expand the tool description to 100-150 characters following the pattern: 'Analyzes user input to identify and rank relevant tools from a registry. Use this when the user's intent requires multiple tools or when the initial query is ambiguous. Returns a scored list of tool names and brief rationales.'
• Document what 'tool selection' means in this context. Is it selecting from a predefined registry? Does it require a tool list as input? What NLP model backs it?
• Document the output schema explicitly: e.g., 'Returns a JSON object with keys: selected_tools (array of {name: string, confidence_score: 0-1, reason: string}), alternative_tools (array), error (string or null).'
• Add error handling guidance: 'If no tools match user intent (all confidence scores < 0.3), return empty selected_tools array and error: "No relevant tools found for this query." The LLM may need to ask for clarification.'
• Consider renaming from 'select_relevant_tools' to 'find_relevant_tools' or 'identify_applicable_tools' to be more explicit about intent. If this is for composition, consider 'suggest_tool_chain' or 'plan_tool_sequence'.

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

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