Server defines 4 tools with explicit schemas and descriptions. Three primary tools (save_mcq_output, save_essay_output, save_summary_output) have well-structured input schemas with nested object validation and clear enum constraints. Descriptions are adequate (80-150 chars) but lack actionable context about when to use each tool vs. alternatives. One deprecated shim tool (save_ai_output) signals API evolution but adds confusion. All tools are WRITE operations with no error recovery guidance. Schemas are complete with type definitions and nested validation, but output schemas are undocumented, LLMs cannot predict what these tools return.

• Document output schema for all tools. Example: 'Returns { success: boolean, jobId: string, savedAt: ISO8601, recordCount: number }'. This enables LLMs to verify success and chain operations.
• Enhance tool descriptions with use-case context. E.g., 'Use save_mcq_output when the AI has generated multiple-choice questions with exactly 4 options and a single correct answer. Use save_essay_output for open-ended prompts with grading rubrics.'
• Add error handling guidance to all tool descriptions. E.g., 'Returns 400 if content.type does not match the tool (e.g., ESSAY content passed to save_mcq_output). Returns 409 if jobId already has saved output, call update_ai_output instead.'
• Remove or clearly deprecate save_ai_output. If it must remain for backward compatibility, add a deprecation notice and sunset date to the description.
• Add per-parameter validation hints. E.g., 'jobId must be a valid UUID v4. materialId must be a valid UUID v4. If either is invalid, the tool returns 400 with the invalid value and the expected format.'