Server has 15 tools with consistent naming (verb_noun pattern) and descriptions present for all. However, parameter schemas lack depth: many parameters missing type constraints, enums, and detailed descriptions. Output schemas are not documented. Error handling guidance is minimal. Tool descriptions average ~80 chars (below 194-char baseline), limiting LLM selection clarity. No evidence of pagination support for list tools. Security patterns (secret injection, permission gates) not visible in provided code.

• Add enum constraints to 'provider' and 'model' parameters. Document valid values or reference list_ducks/list_models as discovery tools. Example: 'provider (string, enum: see list_ducks output), The LLM provider name.'
• Document output schemas for all tools. For 'compare_ducks', specify: {responses: [{provider: string, model: string, response: string, tokens_used: number}], total_cost: number}. This enables LLMs to extract and chain results.
• Expand tool descriptions to 100-150 chars. Clarify use cases: 'duck_council: Iterative multi-provider discussion toward consensus (3+ rounds). Use when you need collaborative refinement. duck_debate: Structured adversarial discussion with assigned positions. Use when you need opposing viewpoints.'
• Add pagination to list_ducks and list_models: limit (1-100, default 20), offset (default 0), and return total_count. Document: 'Returns up to 20 providers per call. Use offset to paginate.'
• Add error recovery hints to tool descriptions. Example for ask_duck: 'If provider not found, call list_ducks() first. If model not available, call list_models(provider) to see options.'

Inferred effective spec: 2025-06-18+.

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