legends-mcp offers 8 tools with clear action-verb naming (suggest, list_legends, summon_legend, etc.) that follow the verb_noun convention well. However, significant quality gaps prevent a higher score: (1) Tool descriptions are present but vary widely in specificity and actionability, some lack context about when to use vs. alternatives (e.g., 'get_legend_context' vs 'get_legend_insight' are not clearly distinguished). (2) Input schemas are visible for all tools in the registration, but parameter descriptions are minimal in many cases. For example, the 'vibe' parameter in list_legends has a description, but many optional parameters lack clear guidance on their impact. (3) Output schemas are NOT documented, there is no structured description of what each tool returns or how fields map to downstream tool inputs, violating the pattern:response-shaper requirement. (4) Error handling is present in the source code (try/catch blocks with error messages) but lacks recovery guidance, errors do not tell the agent what to do next or whether to retry. (5) The tool set shows mild composition concerns: 'get_legend_context' and 'get_legend_insight' perform overlapping operations without clear differentiation, and 'summon_legend' vs 'get_legend_context' have ambiguous boundaries. Tools are READ_ONLY and follow single-responsibility at a high level, but interdependencies are not explicitly documented.

• Document the output schema for each tool. For example, 'list_legends returns an object with fields: legends (array of {id, name, category, description}), total_count (number), and category_filter (string or null). Use this in subsequent tool calls requiring legend IDs.'
• Clarify tool boundaries and selection criteria. Update tool descriptions to explicitly state when to use 'get_legend_context' (comprehensive, for deep dives) vs 'get_legend_insight' (quick, focused on topic). Similarly, clarify 'suggest' (AI-chosen based on message) vs 'auto_match' (AI-ranked by content similarity) vs 'search_legends' (keyword search).
• Enhance parameter descriptions with constraints and format guidance. For example, 'format: one of [full, markdown, system_prompt, frameworks, principles, voice, examples], use "full" for comprehensive context, "system_prompt" for LLM role-play setup' instead of just listing the enum.
• Add error recovery guidance. When a legend_id is not found, return: 'Legend "elon-musk" not found. Did you mean: [elon_musk, elonmusk] or use search_legends() or list_legends().'
• Specify result limits and pagination. Add to list_legends, search_legends, and party_mode: 'Returns up to 20 results by default. Use pagination or limit parameter for larger datasets.'
• Consider merging 'get_legend_context' and 'get_legend_insight' into a single tool with a 'detail_level' enum parameter (quick, detailed, full) or splitting them more clearly by use case with distinct naming (e.g., 'get_legend_full_context' and 'get_legend_quick_insight').