Server has 9 tools with basic schemas and descriptions, but significant gaps limit production readiness. All tools have descriptions (10-100 chars, mostly adequate) and input schemas with types. However, parameter descriptions are sparse or missing context. No output schemas documented. Error handling is absent, tools will fail silently or return raw exceptions. No tool annotations (readOnlyHint/destructiveHint). Naming is clear (verb_noun pattern) but some tools conflate concerns (store_text + search_similar_text in same module). No pagination support despite search tools. Missing validation guidance and recovery hints.

• Add output schemas to all tools. Document response structure: e.g., get_points returns {points: [{id, vector?, metadata}], count}. search_similar_text returns {results: [{id, score, metadata}], total}.
• Implement error handling with recovery hints. E.g., 'Collection not found. Call list_collections() to see available collections.' or 'Invalid filter JSON. Expected valid JSON object with Qdrant filter syntax.'
• Add tool annotations via fastmcp: @mcp.tool(annotations={'readOnlyHint': True}) for read tools, {'destructiveHint': True} for delete_points, {'idempotentHint': True} for upsert_vectors.
• Add pagination to search tools: accept limit (default 10, max 100) and offset/cursor parameters. Return total_count and next_cursor in response. Document in descriptions: 'Returns up to 10 results; use offset for pagination.'
• Expand parameter descriptions with format/constraint guidance. E.g., filter_json: 'JSON string representing Qdrant filter condition (e.g., {"key": "status", "match": {"value": "active"}}). See Qdrant docs for syntax.' metadata: 'Optional dict of key-value pairs (strings/numbers/bools only).'

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

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry