Canvas MCP provides 23 tools with complete input schemas and descriptions, but quality is uneven. All tools follow verb_noun naming convention and every parameter has a type and description. However, descriptions are brief (averaging 40-80 chars) and lack LLM-optimized context like 'when to use this' and 'what it returns'. Output schemas are undocumented, the implementation returns dict[str, Any] with no specification of what fields agents should expect. Error handling is absent from the definitions; no guidance on retryability, user-fixable vs. fatal errors, or recovery paths. The tools cover legitimate read/write operations on Canvas LMS but lack the composition guidance needed for multi-step workflows. Tool grouping is reasonable (assignments, courses, discussions, submissions) but parameter relationships are not documented (e.g., preview_assignment_submission has 'now' and 'caffeinate' together, but the description says 'Illegal with --now', this constraint is buried in the parameter description rather than highlighted in a dependency section).

• Document output schemas for all tools. For each tool, specify: (a) the top-level structure (object/array), (b) field names and types, (c) which fields are guaranteed vs. optional. Example for list_course_assignments: 'Returns: {"assignments": [{"id": string, "name": string, "due_at": string (ISO-8601), "submission_types": string[]}], "total": integer, "next_cursor": string | null}'.
• Add LLM-optimized descriptions (100-200 chars). Include: what the tool does, when to use it instead of similar tools, and what fields the response contains. Example: 'Retrieve a single assignment with full details including due date, submission types, and current user submission. Use this after list_course_assignments to get rubric or submission details.'
• Convert free-form string parameters to enums. Add JSON Schema 'enum' array to 'bucket' (upcoming, overdue, past) and 'submission_type' (online_upload, online_text_entry, etc.). This prevents hallucinated values.
• Document pagination explicitly. For all list_* tools, specify whether the response includes 'total_count', 'next_cursor', 'next_offset', or is limited to the 'limit' parameter. Example: 'Pagination: Returns up to 'limit' items. Include 'offset' or 'cursor' (if returned) to fetch the next page.'
• Clarify parameter dependencies. For preview_assignment_submission, add a 'Constraints' section to the description: 'Mutually exclusive: either 'now' (submit immediately) OR 'minutes_before_due' / 'submit_at' (schedule). If 'now' is true, 'caffeinate' is invalid.'