Server has 4 well-named tools with clear descriptions (avg 150 chars, within baseline 194). All tools have input schemas with typed parameters and descriptions. However, output schemas are not documented, LLMs cannot predict return structure. Parameter descriptions lack format constraints (e.g., UUID format, enum values for status/type). Error handling and recovery guidance are absent. No tool annotations (readOnlyHint/destructiveHint). Composition is sound: tools chain logically (search → add → update). Security concern: embedding vectors are exposed as parameters, risking token leakage if logged.

• Document output schema for each tool. E.g., get_notes returns {id: uuid, topic: string, content: string, type: enum, status: enum, project: string, created_at: ISO8601, similarity: float, rrf_score: float}. Include pagination (total_count, next_cursor).
• Convert 'type' and 'status' parameters to enums with explicit allowed values. E.g., type: enum[decision|insight|idea|task|journal|to_study|contact_note|question]. Add format: uuid to 'id' parameter.
• Move embedding vector generation server-side. Remove 'query_embedding' and 'embedding' from tool parameters. Accept 'query_text' only; compute embedding internally using OpenAI API. Prevents token leakage.
• Add tool annotations: mark add_note and update_status with destructiveHint=true; mark get_notes and get_topics with readOnlyHint=true. Helps agents reason about side effects.
• Add error handling guidance to descriptions. E.g., 'add_note: Returns 400 if type not in [decision|insight|...]. Returns 409 if related_ids reference non-existent notes. Embedding generation is async; check status via get_notes after 5s.' Include retryability classification.