Critical failures across all dimensions. The server dynamically generates tools from OpenAPI specs with names like 'API-{operationId}' (truncated to 64 chars). Tool definitions are inferred from OpenAPI parsing, not explicitly registered in code. The single tool entry shows an empty properties object with additionalProperties:true, no actual schema is visible. Description is generic boilerplate about 'dynamically generated tools' with no actionable guidance for LLMs. No parameter descriptions, no output schema documentation, no error handling guidance. This violates core patterns: tool naming (no verb prefix, opaque operationId names), tool description (generic, under 50 chars of substance), parameter documentation (completely absent), and schema completeness (empty/dynamic).

• Explicitly register each OpenAPI operation as a separate MCP tool with a proper verb_noun name (e.g., 'list_databases', 'create_page', 'update_block') instead of opaque 'API-{operationId}' names. This requires parsing the OpenAPI spec at server startup and calling server.setTools() with explicit tool definitions.
• For each tool, generate a description that answers: What does it do? When should the LLM call it? What does it return? Example: 'list_databases: Retrieve all databases accessible to the authenticated user. Returns database ID, title, and icon. Call this first to discover available databases before querying or updating them.'
• Generate parameter descriptions from OpenAPI parameter descriptions and schemas. For each parameter, include: type, format, constraints (min/max, enum values), and whether it is required. Example: 'database_id (string, required): The UUID of the target database. Use list_databases() to discover available IDs.'
• Document output schema for each tool. Parse the OpenAPI response schema and generate a summary of returned fields, types, and meaning. Include pagination info (limit, offset, total_count) if applicable.
• Implement error handling that returns actionable recovery guidance. Catch HTTP errors and map them to LLM-friendly messages: 'Database not found. Call list_databases() to see available options.' instead of raw 500 errors.