Discord MCP presents a dispatcher pattern with 6 tools of highly variable quality. Three dispatcher tools (mcp_tools_search, mcp_tools_read, mcp_tools_write, mcp_tools_destructive) have strong naming and good descriptions but invisible input schemas (inferred from JSON samples only). The build_discord_server tool is mentioned but has no visible schema or detailed parameters. The components_v2_preview tool has a minimal schema. Tool descriptions are informative but several lack actionable parameter guidance. Critically, actual Discord operation tool schemas are not visible in the provided code, only dispatcher interfaces are shown. This makes it impossible to assess the quality of the underlying tool catalog. The server appears designed as a catalog/dispatcher rather than direct tool implementation, which creates a scoring problem: we cannot verify the schemas, parameters, and error handling of the actual Discord tools being dispatched.

• Make build_discord_server a documented tool with explicit schema. Currently it has only a name and risk label. Add inputSchema with parameter definitions, descriptions, type constraints, and expected return structure.
• For dispatcher tools, replace the generic 'args' parameter with explicit schema validation. Either enumerate the possible tool signatures or generate validated subschemas based on the catalog discovery.
• Provide actual example schemas or links to the underlying tool catalog in the source code or documentation. The evaluation cannot verify Discord operation tool quality without seeing their definitions.
• Add return schema documentation for mcp_tools_search showing the exact structure of compact vs full results (e.g. {name, description, contract?}, where contract includes inputSchema, outputSchema, riskLevel).
• Document when to use mcp_tools_search discovery vs direct tool dispatch. Add a workflow guide: 'If unsure which tool to call, use search first. For known tools, use read/write/destructive directly.'
• Add error recovery guidance to all tools. Example: 'If tool not found error, call mcp_tools_search again with a broader query.' Currently error handling is implicit.
• Specify pagination and limits for mcp_tools_search. Current 'limit' defaults to 8 and caps at 20, document this in the description and explain why (to prevent overwhelming LLM context).
• For components_v2_preview, add a full example of the ASCII output format in the description or documentation. Describe multi-line structure, any special characters, and maximum output size.

Inferred effective spec: 2025-06-18+.

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