Critical failure across all 14 tools. Every tool has EMPTY input schemas (no properties, no required fields) and generic 1-2 sentence descriptions. Tools are named with 'API-' prefix (non-standard verb_noun pattern). No parameter descriptions, no output schemas documented, no error handling guidance. This is a proxy server that exposes raw OpenAPI operations without MCP-specific optimization.

• Rename all tools from 'API-verb' to 'verb_noun' format: 'create_database', 'query_database', 'retrieve_page', 'delete_block', 'search_notion'. This matches production baselines and helps LLMs parse intent from names alone.
• Expand each description to 100-200 characters with: (1) what the tool does, (2) when to use it vs similar tools, (3) key parameters, (4) what it returns. Example: 'Query a Notion database by filter, sort, or search. Returns paginated results with page IDs for chaining to retrieve_page. Use after create_database or when you have a database_id.'
• Add full input schemas with typed parameters and descriptions. For API-queryDatabase: add 'database_id' (string, required), 'filter' (object, optional), 'sorts' (array, optional), 'page_size' (integer 1-100, default 20), 'start_cursor' (string, optional for pagination). Every parameter must have a description explaining format and constraints.
• Document output schemas for each tool. Example for API-queryDatabase: return {results: [{page_id, title, created_time, last_edited_time}], next_cursor, has_more}. Include all fields needed for downstream tool calls (e.g., page_id for retrieve_page).
• Add error handling and recovery guidance. For destructive tools (delete_block, update_database), include: 'This operation is irreversible. Verify the ID before calling. If you only have a name, call search_notion() first.' For read tools, add: 'If not found, try search_notion() with a partial name or filter.'

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

• Relies on Stateful initialize / Mcp-Session-Id (removed; protocol is stateless) - make each request self-contained