OpenKLAS exposes 14 well-intentioned tools via fastapi-mcp with HTTP transport. Tool names follow verb_noun patterns consistently (login, logout, get_me, list_eclass_lectures, etc.), which is a strength. Descriptions are present for all tools and range 80 - 280 characters, meeting the 10 - 1024 character baseline. However, several critical gaps reduce quality: (1) Input schemas are visible and properly typed for most tools, but lack enum constraints on enums (semester accepts '1' or '2', should be declared as enum, not free-form string). (2) Descriptions lack LLM-optimized guidance: they do not state WHEN to use each tool vs. alternatives, dependency hints ('call X first to get Y'), or recovery paths for errors. (3) Output schemas are entirely undocumented, the server returns responses but provides no description of the return structure, forcing LLMs to infer fields. (4) Error handling is minimal, no tool documents what error conditions are possible, what they mean, or how to recover. (5) Parameter descriptions are present but terse and sometimes ambiguous (e.g., 'year' param lacks default behavior if omitted, despite description saying 'Defaults to current'). (6) No tool declares permissions (read:email, write:calendar) or audit scope. (7) Composition is reasonable (14 focused tools), but some miss natural chaining IDs (e.g., summarize_eclass_lecture returns nothing immediately, status must be polled separately, should return a job_id for follow-up).

• Document output schemas for all tools. For each tool, add a 'Returns' section describing field names, types, and nesting. Example: 'Returns {lectures: [{lecture_id: string, title: string, video_url: string, date: ISO8601}], total_count: integer}'.
• Declare enums for constrained parameters. Change semester parameter from free-form string to enum(['1', '2']) and year to integer with description 'Academic year (2020-2099)'. Use JSON Schema enum keyword.
• Add error handling documentation. For each tool, document expected errors: 'Errors: 401 Unauthorized (invalid token, call login), 404 Not Found (subject not enrolled), 500 Internal Server (API timeout). Recovery: check credentials, verify subject_code, retry after 5s.'
• Standardize parameter names across related tools. Use 'attach_file_id' consistently (not 'atch_file_id'), 'weekly_seq' not 'weeklyseq'. Verify that response field names match parameter names in downstream tools.
• Implement job ID pattern for async operations. Have summarize_eclass_lecture return {job_id: string, status: 'queued'}. Update eclass_summarize_status to accept job_id parameter. Support concurrent requests.
• Add tool annotations to distinguish safe operations from destructive ones. Use inputDescription for login/logout with explicit permission scope hints (e.g., 'This endpoint writes user session state; requires valid student credentials').
• Clarify Bearer token format in logout. Specify: 'Authorization header value must be in format "Bearer <token>". Pass only the token value (without 'Bearer' prefix) in the credentials parameter.'

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

• Relies on Dynamic Client Registration (deprecated) - use Client ID Metadata Documents (CIMD)